Breast Cancer Classification Using Breast Ultrasound Images with a Hybrid of Transfer Learning and Bayesian-Optimized Fast Learning Network

Breast Cancer Classification Using Breast Ultrasound Images with a Hybrid of Transfer Learning and Bayesian-Optimized Fast Learning Network | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Breast Cancer Classification Using Breast Ultrasound Images with a Hybrid of Transfer Learning and Bayesian-Optimized Fast Learning Network Emmanuel Ahishakiye, Fredrick Kanobe This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-5333695/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 29 May, 2025 Read the published version in Discover Artificial Intelligence → Version 1 posted 17 You are reading this latest preprint version Abstract Background Breast cancer remains the most frequent cancer diagnosed in females, resulting in high mortality rates worldwide. Approximately 2.3 million cases are diagnosed annually. If it is detected at an early stage, the rate of survival is significantly improved; therefore, there is an urgent need for techniques that can be used for its effective diagnosis. Method The study aimed to present a hybrid model for breast cancer classification by employing DenseNet201 as a feature extractor and Bayesian-Optimized Fast Learning Network (FLN) as a classifier. The pre-trained DenseNet201 extracts high-quality features from breast ultrasound images on large datasets, which get classified through an FLN optimized using Bayesian techniques for hyperparameter tuning. Results The model performed well by achieving an accuracy of 96.79%, 94.71% F1 score, 96.81% precision, and 93.48% recall, while the AUC for benign, malignant, and normal cases was found to be 0.96, 0.95, and 0.98, respectively. Cross-entropy loss metrics further validated the model on its robust training and validation. Conclusion There is a great potential that this proposed model could enhance breast cancer diagnosis. This indeed is a reliable and efficient clinical solution for application. Breast Cancer Machine learning Deep learning Fast Learning Network Bayesian Optimization Figures Figure 1 Figure 2 Figure 3 Figure 4 1. Background of the Study Breast cancer is the most diagnosed cancer in women worldwide accounting for 11.7% of all new cases. An approximate number of 670,000 deaths due to the disease were recorded in 2022 alone [ 1 ]. In 2022 alone, 2.3 million new cases of breast cancer were diagnosed, making this cancer type a big concern for public health and also the leading cause of death from cancer in females around the world. It is very important to diagnose cancer at an early stage, as it can remarkably improve the outcome of a patient. For instance, the five-year survival rate for localized and early-stage breast cancer is 99%, while it is only 28% when the diagnosed cases reach a farther stage onward [ 2 ]. Mammography, ultrasound, and MRI are the conventional medical imaging modalities employed in detecting and diagnosing breast cancer [ 3 ]. However, accurately interpreting these images can be challenging, even for experienced radiologists [ 4 ]. The sensitivity ranged from 25 − 58% for mammography, 33 − 52% for ultrasound, 48 − 67 for mammography plus ultrasound, and 71 − 100% for MRI in women with dense breast tissue, leading to missed detections and false negatives [ 5 ]. Therefore, automatic schemes are highly wanted to assist radiologists in correctly detecting the type of breast cancer. Medical image analysis has been greatly improved due to the advances in machine learning and deep learning methods [ 6 ] [ 7 ]. These methods provide automated and objective support to classify breast cancer, with minimal or no dependence on manual feature extraction [ 8 ]. Among deep learning methods, CNN has been very successful in the classification tasks of breast cancer by automatically learning complex patterns from raw image data. For example, some CNN-based methods applied to breast ultrasound images have reached a very good performance in classifying positive versus negative cases [ 9 ]. However, deep learning models are confronted with such issues as requiring a lot of labeled data, running overfitting risks in the case of small datasets, and high computational loadings [ 10 ] [ 11 ]. One of the various strategies that have been employed to resolve this is transfer learning [ 12 ]. Pre-trained models on large datasets like ImageNet have enabled knowledge acquired from general image classification tasks to be transferred for the analysis of medical images, thereby avoiding the need for labeled data in great quantity. Applications of transfer learning in breast cancer detection have been made using VGG16, ResNet, and DenseNet models and reached excellent classification accuracies [ 13 ]. Among them, DenseNet201 achieved the best performance in feature extraction due to its architecture wherein all the layers are connected densely to allow improved gradient flow and propagation of features [ 14 ]. While transfer learning enhances model performance, another critical challenge in practice is the optimization of hyperparameters of the learning model, such as the number of neurons, the learning rate, or the dropout rate [ 15 ]. Traditional methods, such as grid search and random search, are rather inefficient for the exploration of high-dimensional hyperparameter spaces of deep learning models [ 16 ]. Bayesian optimization, on the other hand, is a more efficient alternative that uses probabilistic models to guide the search for optimal hyperparameters [ 17 ]. This approach has outperformed the traditional methods in fine-tuning complex models. Bayesian optimization works particularly well for lightweight models such as the Fast Learning Network, which, if hyperparameters are carefully optimized, can give high accuracy with fast training speeds. Integration of transfer learning and Bayesian-optimized FLN is a very promising approach toward detection and classification in the case of breast cancer. This hybrid approach features extraction from breast ultrasound images, which uses a deep pretrained model like DenseNet201. These are then fed into an FLN model optimized using Bayesian techniques for finding the best hyperparameters. It was hypothesized that the proposed model would result in an improvement in the classification performance with limited data and model complexity by combining the strengths of transfer learning in effective feature extraction with Bayesian optimization in efficient model tuning within one hybrid framework. The proposed model is efficient and accurate for the classification of breast cancer and can potentially improve clinical decision-making and patient care. This study has been developed for SDG 3: " Ensure healthy lives and promote well-being for all at all ages ," within the theme of artificial intelligence for development. 2. Literature Review 2.1 Machine Learning in Breast Cancer Classification The study [ 18 ] proposed a three-step image processing approach that included RGB fusion, region of interest highlighting, and speckle noise filtering utilizing a block-matching three-dimensional filtering technique. This approach improves performance and broadens the generalization of deep learning models. Three datasets were employed in the study for transfer learning: BUSI (780 images), Dataset B (162 images), and KAIMRC (5693 images) using a deep learning model (VGG19). The model with the suggested preprocessing step outperformed the model without preprocessing for each dataset when tested using a fivefold cross-validation procedure on the BUSI and KAIMRC datasets. The deep learning classification model for breast cancer performs better with the proposed image processing method. The study (Uddin et al., 2023) employed the Wisconsin Breast Cancer Dataset (WBCD) as a training set from the UCI machine learning library to examine the effectiveness of the different machine learning techniques. To compare and analyze breast cancer into benign and malignant tumors, various machine learning classifiers have been used, including support vector machines (SVM), Random Forests (RF), K-nearest neighbors (K-NN), Decision Trees (DT), Naïve Bayes (NB), Logistic Regressions (LR), AdaBoost (AB), Gradient Boosting (GB), Multi-layer Perceptrons (MLP), Nearest Cluster Classifiers (NCC), and voting classifiers (VC). The research reveals that the voting classifier has the lowest mistake rate and the best accuracy, at 98.77%. The study [ 19 ] proposed a deep learning-based technique that uses an Inception-v3 convolutional neural network to fine-tune images of breast tissue stained with Hematoxylin and Eosin (H&E). It is necessary to categorize these images into four groups: normal tissue, benign lesions, in situ cancer, and invasive carcinoma. Through majority vote over the nuclear classes, the class of the overall image is established. The findings showed a 93% accuracy rate for non-cancer (normal or benign) versus malignant (in situ or invasive carcinoma) and an average accuracy of 85% across the four classes. The study [ 20 ] introduced a neural network called the Multi-modal Transformer (MMT), which uses ultrasound and mammography in concert to identify people with cancer and calculate the probability of developing cancer in those who are currently cancer-free. MMT uses self-attention to aggregate multimodal data and compares the present examinations to previous images to monitor changes in tissue over time. MMT surpasses robust uni-modal baselines with an AUROC of 0.943 in detecting pre-existing tumors after being trained on 1.3 million tests. With an AUROC of 0.826 for 5-year risk prediction, MMT outperforms earlier mammography-based risk models. The study [ 21 ] proposed a computer-aided diagnosis (CAD) system that is capable of producing an optimal algorithm on its own. 13 of the 185 available features are used to train machine learning. Five machine-learning classifiers were employed to distinguish between benign and malignant tumors. Based on a machine learning classifier, the experimental findings demonstrated Bayesian optimization using a tree-structured Parzen estimator for 10-fold cross-validation. Outperforming the other four classifiers, the LightGBM classifier achieves 99.86% accuracy, 100.0% precision, 99.60% recall, and 99.80% for the FI score. The study [ 22 ] examined the different Deep-learning models for categorizing photos of breast cancer histology. For the ImageNet database, models such as ResNeXt, Dual Path Net, SENet, and NASNet have been found to produce the most advanced results. The study also looked at the Inception-ResNet-V2 architecture, which produced better comparison results and produced the greatest results for binary and eight classifications. The study [ 23 ] provided a thorough literature assessment of deep learning-based techniques for detecting breast cancer, which can help researchers and practitioners comprehend the difficulties and emerging trends in the field. With an emphasis on genomic and histopathological imaging data, various deep learning-based techniques for breast cancer identification are specifically examined. The review's findings showed that the Convolutional Neural Network (CNN) is the most widely used and accurate model for detecting breast cancer and that the most widely used technique for assessing performance is accuracy measures. To enhance the detection of breast cancer classification, the study [ 24 ] proposed a new deep model to classify breast cancer. It was motivated by two cutting-edge deep networks, GoogLeNet, and residual block, and developed several new characteristics. The accuracy of the suggested model on ultrasound and breast histopathology images was 93% and 95%, respectively. The study [ 25 ] proposed a Meta-Learning Ensemble Method for Breast Cancer Classification Using Convolution Neural Networks. Using a meta-learning framework, the proposed approach combined several CNN models, such as InceptionV3, ResNet50, and DenseNet121, to increase generalization and achieve a 90% accuracy rate, especially when identifying malignant tumors. The study [ 26 ] used a primary dataset to assess and compare five distinct machine learning techniques' classification accuracy, precision, recall, and F1 scores. We employed five distinct supervised machine learning methods to attain the best outcomes on our dataset: logistic regression, naive Bayes, decision tree, random forest, and XGBoost. This study's final evaluation revealed that XGBoost had the best model accuracy, at 97%. To identify breast cancer on digital breast tomosynthesis (DBT) images, the study [ 27 ] suggested a method for creating an effective deep neural network model that takes into account context from nearby image segments. The two 3D models outperformed the per-section baseline model in classification on the test set of 655 DBT trials. When compared with the single-DBT-section baseline, the suggested transformer-based model showed a significant increase in AUC (0.88 vs 0.91, P = .002), sensitivity (81.0% vs 87.7%, P = .006), and specificity (80.5% vs 86.4%, P < .001) at clinically important operating points. While showing comparable classification results, the transformer-based model only utilized 25% of the floating-point operations per second that the 3D convolution model did. The study [ 28 ] presented a self-attention Vision Transformer model designed especially for histological image classification of breast cancer. Pre-training, dimension scaling, data augmentation, color normalization techniques, patch overlap, and patch size configurations are some of the training strategies and configurations that we analyze to assess their effects on the performance of histology picture categorization. An accuracy rate of 0.91, 0.74, and 0.92 on the BACH, BRACS, and AIDPATH datasets are attained by the model built from a transformer pretrained on ImageNet. Pre-training on big generic datasets such as ImageNet appears to be beneficial, according to the results. 3. Materials and Methods 3.1 Data Description Datasets [ 29 ] were collected at Baheya Hospital. Ultrasound (US) images are usually grayscale. The data was collected from women between the ages of 25 and 75 years old. The number of patients is 600 female patients. The dataset consists of 780 images with an average image size of 500x500 pixels. The US dataset is divided into three categories: benign, malignant, and normal. There were 1100 images gathered at the start and following the data preprocessing, only 780 images remained. LOGIQ Agile and LOGIQ E9 ultrasound systems are the tools utilized in the scanning procedure. Typically, these devices are employed in high-quality imaging for cardiac, vascular, and radiological applications. They generate images with a resolution of 1280 x 1024. 3.2 Data preprocessing Data preprocessing was done on all the Pap smear images to maintain uniform quality and relevance for model training. This was done to improve the diversity of the dataset, data augmentation approaches can be employed, including rotation, flipping, and scaling. This increases the generalization capability of the models. The pixel values are finally normalized into a standard range from [0, 1], to ensure that the nature of the input provided to machine learning algorithms is always uniform. These preprocessing steps attempt to improve the quality of images and make the dataset suitable for the accurate classification of cervical cancer. 3.3 The Proposed Model The proposed model for the detection and classification of breast cancer combines the strength of DenseNet201 in feature extraction with a Bayesian-optimized Fast Learning Network (FLN) for classification. DenseNet201 serves as a robust feature extractor by transferring learned features to breast cancer detection. Thereafter, features are extracted and a light Fully Connected FLN is used for classification. The fully connected layers are optimized with Bayesian optimization to fine-tune the hyperparameters. The finetuned hyperparameters are learning rate, number of neurons, and dropouts. This further assists in the improvement of the model generalization capability with amplified performance. The model then yields class probabilities and can detect benign, malignant, and normal. The proposed model was evaluated using various performance metrics including accuracy, precision, recall, F1-score, and AUC. Besides that, confusion matrices and AUC curves were also employed to evaluate the performance of this model. The proposed approach of transfer learning along with Bayesian optimization aims at high accuracy with reliable classification in breast cancer detection. 3.4 The Pseudocode of the Proposed Model The Pseudocode describes the model, which involves the following steps: a. Pre-processing of images of breast cancers; b. Loading the DenseNet201 pre-trained feature extractor, from which it removes the last classification layers. It is followed by a Fast Learning Network - FLN, containing two fully connected layers optimized with Bayesian optimization in tuning hyper-parameters such as learning rate and dropout. These class probabilities, benign, malignant, and normal are output through a softmax layer at the end of the FLN. Then we compile the model using the Adam optimizer and categorical cross-entropy loss. We test the performance of the model on accuracy, F1-score, precision, recall, and AUC together with the confusion matrix and AUC curves. 3.5 Model Evaluation The proposed models were assessed using several metrics: accuracy, cross-entropy loss, precision, recall, F1-score, and loss function. Accuracy measures the ratio of correctly predicted outcomes to the total number of predictions [ 30 ]. It was chosen for this task because it provides a quick assessment of a model's performance and is particularly effective for binary classification problems. Also, the cross-entropy metric measures the model's error and dissimilarity between predicted and actual values. The performance metrics are calculated using the formulas below: $$\:Accuracy=\:\frac{TP+TN}{TP+TN+FP+FN}\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\left(1\right)$$ $$\:F1-Score=\:\frac{2TP}{2TP+FP+FN}\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\left(2\right)$$ $$\:Precision=\:\frac{TP}{TP+FP}\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\left(3\right)$$ $$\:Recall,\:Sensitivity=\:\frac{TP}{TP+FN}\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\left(4\right)$$ $$\:cross\:entropy\:loss,\:H\left(y,\:\stackrel{´}{y}\right)=-\sum\:_{i=1}^{n}{y}_{i}\text{l}\text{o}\text{g}(\stackrel{´}{{y}_{i}}\left)\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\right(5)$$ where \(\:TP,\:TN,\:FP,\:and\:FN\) stand for true positive, true negative, false positive, and false negative, respectively. Also, \(\:y\) is the true probability distribution (the one-hot encoded true labels), \(\:\stackrel{´}{y}\) is the predicted probability distribution (the model’s predicted probabilities for each class), \(\:n\) is the number of classes in the classification problem, and \(\:log\) is the natural logarithm function. The proposed model was implemented using Python 3.12.4, TensorFlow 2.16.0, Keras 3.4.0, scikit-learn (sklearn) 1.5.0, NumPy 2.0, Pandas 2.2.2 and Matplotlib 3.9.1. Also, a Windows 11 laptop, Intel Core I7 processor of 2.50 GHz, and 16GB RAM with 2GB of NVIDIA GeForce MX150 graphical processing unit (GPU) were used. 4. Results 4.1 Comparative Performance of the Pre-Trained Transfer Learning Models To determine the best pre-trained transfer learning model to integrate with the optimized FLN, we did a performance evaluation of some of the existing transfer learning models. In this work, Table 1 displays a few of the pre-trained transfer learning models that have been refined on the dataset used for cervical cancer. Because these models have already been trained, they can take advantage of feature representations that have been acquired from a wide range of different and sizable datasets. The last layers of each model were changed, and the cervical cancer dataset was used for additional retraining so that the learned characteristics of the model could adjust to this particular classification job. Changing the model parameters, particularly the upper layers, while maintaining fixed or low learning rates for the earlier layers that identify more general features like edges and textures is known as fine-tuning. Each model underwent five-fold cross-validation to evaluate its generalization on unknown data, and the results were recorded for training accuracy, validation accuracy, training loss, and validation loss. By fine-tuning, the models were able to focus on detecting features that are important for breast cancer diagnosis. DenseNet201 was chosen as the study's preferred model since it was the overall best in performance. Table 1 Performance of the Pre-Trained Models Model Training Accuracy (%) Validation Accuracy (%) Training Loss Validation loss VGG16 95.54 92.75 0.82 1.97 VGG19 95.36 91.14 0.89 1.95 ResNet50 95.70 94.35 0.66 2.01 InceptionV3 97.50 95.62 0.45 1.99 DenseNet201 97.93 96.54 0.31 1.83 MobileNetV3 97.15 95.62 0.22 0.24 4.2 Bayesian Optimization of Hyperparameters Results for the optimal values of the various tunings regarding FLN are highlighted in Table 2 , conducted through Bayesian optimization. In this model, the number of 471 hidden neurons is a reasonable balance between learning capability and model complexity. The learning rate is 0.000506, way below 1.0, which allows gradual weight updates, not overshooting, and makes the training process stable for more accuracy. The 44.18% dropout is a technique that reduces overfitting by randomly ignoring neurons during training to enhance the generalization capability of the model. Finally, Adam has been chosen as an optimizer because, with it, adaptive learning rate and momentum are possible, which allows the model performance to converge faster. Table 2 Best Hyperparameters Parameter value Hidden Neurons 471 Learning Rate 0.0005064479220813515 Dropout Rate 0.44180424373490723 Optimizer Adam 4.3 Performance on Accuracy, F1 score, precision, and Recall The proposed model shows very good performance compared to the models in the literature as shown in Table 3 . It has high accuracy of 96.79%, F1 score of 94.71%, precision of 96.81%, and recall of 93.48%. Thus, the model appears competitive since, besides the advantage in accuracy for several models like those by [ 19 ] and [ 25 ], it gives a more balanced set of metrics. Although this does not quite match the near-perfect performance seen in [ 21 ], with an accuracy of 99.86% and perfect precision, the proposed model continues to show very good precision and recall, hence reliable and effective classification of cases of breast cancer. precision of 96.81% indeed depicts a strong ability to minimize false positives, hence making it a robust choice; however, there is still room for further optimization toward the highest-performing models. Table 3 Performance of the model on performance metrics Model Accuracy (%) F1 score (%) Precision (%) Recall (%) [ 21 ] 99.86 99.80 100.00 99.60 [ 31 ] 98.77 - - - [ 19 ] 85.00 - - - [ 24 ] 95.00 - - - [ 25 ] 90.00 90.00 94.00 95.00 [ 28 ] 74.00 - - - Proposed 96.79 94.71 96.81 93.48 4.4 Performance on Confusion Matrix The results of the confusion matrix proposed model are shown in Fig. 2 . The proposed model had a true positive of 165 for Benign and false positive of 11 and a false negative of 7, then a true positive of 72 for malignant with a false negative of 14 and a false positive of 7, and finally a true positive of 39 for normal cases with the false negative of 2 and false positive of 1. The model performs very well for the benign and normal classes, with a high number of true positives and relatively low false positives and negatives. 4.4 Performance on AUC for each Class The AUC is a measure of the performance of the ability of the model in differentiating classes. AUC ranges between 0 and 1, with values close to 1 indicating better performance. The proposed model obtained the values of AUC for classes as benign (0.96), malignant (0.95), and normal (0.98) demonstrating that the model did a great a great job in classifying all three classes of breast cancer with the most excellent ability to correctly identify the normal cases. The relatively lower AUC for the malignant class may be indicative of potential improvements but overall, the model demonstrated great performance. The AUC of the proposed model is shown in Fig. 3 . 4.5 Results on Cross-Entropy Loss The Cross-entropy loss quantifies how distant the predicted probability distribution is from the true distribution of target classes. A lower cross-entropy loss indicates that predictions are closer to real class labels. When it has a value near zero, that will mean very good accuracy; with higher values, the accuracy will be bad. During training time, a decreasing cross-entropy loss is an indication of a well-learning model that slowly gets better in its predictions. Figure 4 shows that with the increase in iteration, cross-entropy loss decreased both in training and validation, proving our proposed model's effectiveness. 5. Discussion of the Results The proposed model for the classification of breast cancer was based on DenseNet201 as the base feature extractor with a Bayesian-Optimized Fast Learning Network. Overall, its performance was excellent in all metrics. The quantitative comparison of the results of the proposed method with the state-of-the-art pre-trained models is shown below in Table 2 ; the best results corresponded to DenseNet201 with a validation accuracy of 96.54%, while in other implemented models, VGG16 had a high performance of 92.75% and the closest, ResNet50 achieved 94.35%. It also means that DenseNet201 has layered connectivity and uses its parameters effectively to emphasize the relevant features for diagnosing breast cancer; it is therefore well-motivated to be integrated with FLN. The fine-tuning enabled the model to adapt the pre-trained feature representations to the specifics of the cervical cancer dataset, thus enabling improvement in both classification accuracy and generalization. The optimized FLN’s hyperparameters, as presented in Table 2 , were crucial in further refining the model’s performance. The Bayesian optimization yielded a learning rate of 0.000506 and a dropout rate of 44.18%, with 471 hidden neurons to strike a balance between learning ability and model complexity. The chosen Adam optimizer enabled efficient convergence by adjusting learning rates adaptively, which facilitated the model’s ability to stabilize quickly while minimizing overfitting. These adjustments are reflected in the proposed model’s high accuracy of 96.79% and a strong F1 score of 94.71%, indicating effective handling of both false positives and false negatives, making it a robust solution for breast cancer classification. Hence, this proposed model is competitive with the literature's existing models. Although not near-perfect performance, as seen by [ 21 ], with 99.86% accuracy and perfect precision, the proposed model still outperforms models like [ 25 ] and [ 24 ] in terms of both accuracy and precision. The proposed model had a precision of 96.81%, showing good performance in terms of reducing false positives, a quite important ability within a medical diagnosis environment while its recall of 93.48% shows very reliable sensitivity in depicting the actual positive cases. The robustness of the proposed model is further confirmed by the confusion matrix results, represented in Fig. 2 , showing the effectiveness of the model, especially in the case of benign and normal cases. In this respect, with true positives of 165 for benign and 39 for normal, while having relatively low false positives and false negatives, the model has shown high sensitivity and specificity for these classes. While the malignant class has relatively somewhat higher misclassification rates, it shows 14 false negatives. This means additional tuning will be required to boost recall for malignant instances, by either augmenting training data on malignant features or using class-specific weighting during training itself. More so, from the perspective of AUC scores, the proposed model has the best class discrimination ability, with AUC values of 0.96 for benign, 0.95 for malignant, and 0.98 for normal. A high AUC in a normal class means outstanding ability in correctly identifying normal cases, whereas a little lower AUC value observed in malignant, 0.95, indicates that there may be some difficulties in distinguishing the malignant case from the rest. The strong overall performance in AUC reinforces the robustness of the model, but its performance could be further improved with targeted improvements in malignant classification. The results of cross-entropy loss prove that there has been effective learning throughout the training, clearly seen by the drop in both training and validation loss at each iteration. This downward slope of cross-entropy loss confirms that with time increasing, the model was aligning its predictions toward true labels by gradually increasing the accuracy. Fairly low cross-entropy loss at the end of training pinpoints the model's reliability and hence supports the conclusion that the proposed approach will be accurate and stable. Therefore, the performance of the proposed model for the breast cancer classification is quite rounded and effectively balanced between accuracy, precision, and recall. Feature extraction through DenseNet201 coupled with optimized FLN thus performs reliable classification among benign, malignant, and normal classes. Although this model performs very well, especially in reducing false positives, the enhancement in sensitivity concerning malignant cases will easily bring it closer to the model performance of the top results reported in the literature. 6. Conclusion This paper proposes a hybrid model that involves DenseNet201 and a Bayesian optimization-based Fast Learning Network for efficient detection and classification of BC. The results showed that the proposed model leveraged pre-trained feature extraction afforded by DenseNet201, augmented by flexibility for fine-tuning by Bayesian-optimized FLN, achieved robust results on all metrics of accuracy at 96.79%, F1 score at 94.71%, precision at 96.81%, and recall at 93.48%. Besides, the high values of AUC for benign, malignant, and normal classes prove this model is reliable in classifying categories of interest in breast cancer. Although such results show competitiveness compared with state-of-the-art models, slight improvements might boost sensitivity, especially in the detection of malignancy. This concept has shown huge possibility in the combined application of transfer learning with Bayesian optimization toward improving the accuracy of classification in medical imaging for timely diagnosis in the clinical setting. Future studies may consider improvements in data augmentation and ensemble learning to increase model sensitivity and overall accuracy. Declarations Author contributions: Emmanuel Ahishakiye (EA) and Fredrick Kanobe (FK) designed the study. EA wrote the manuscript. FK edited and extensively reviewed the manuscript. All the authors approved the manuscript. Funding: We acknowledge the financial support from Kyambogo University's competitive grants. Data Availability: The data utilized in this study is sourced from the Dataset of breast ultrasound images,” Data Br. , vol. 28, p. 104863, 2020" which is publicly accessible at the following URL: https://www.kaggle.com/datasets/aryashah2k/breast-ultrasound-images-dataset. Accessed 10-October-2024. doi: 10.1016/j.dib.2019.104863. Code availability: Not applicable. Competing interests The authors declare that they have no competing interests. References World Health Organization, “Breast cancer.” [Online]. Available: https://www.who.int/news-room/fact-sheets/detail/breast-cancer American Cancer Society, “Breast Cancer Facts & Figures.” Accessed: Oct. 25, 2024. [Online]. Available: https://www.cancer.org/research/cancer-facts-statistics/breast-cancer-facts-figures.html E. Abu Abeelh and Z. AbuAbeileh, “Comparative Effectiveness of Mammography, Ultrasound, and MRI in the Detection of Breast Carcinoma in Dense Breast Tissue: A Systematic Review,” Cureus, vol. 16, no. 4, 2024, doi: 10.7759/cureus.59054 . K. Loizidou, R. Elia, and C. Pitris, “Computer-aided breast cancer detection and classification in mammography: A comprehensive review,” Comput. Biol. Med. , vol. 153, no. November 2022, p. 106554, 2023, doi: 10.1016/j.compbiomed.2023.106554 . F. Sardanelli, V. Magni, G. Rossini, F. Kilburn-Toppin, N. A. Healy, and F. J. Gilbert, “The paradox of MRI for breast cancer screening: high-risk and dense breasts-available evidence and current practice,” Insights Imaging, vol. 15, no. 1, 2024, doi: 10.1186/s13244-024-01653-4 . M. Li, Y. Jiang, Y. Zhang, and H. Zhu, “Medical image analysis using deep learning algorithms,” Front. Public Heal. , vol. 11, no. November, pp. 1–28, 2023, doi: 10.3389/fpubh.2023.1273253 . P. K. Mall et al. , “A comprehensive review of deep neural networks for medical image processing: Recent developments and future opportunities,” Healthc. Anal., vol. 4, no. June, p. 100216, 2023, doi: 10.1016/j.health.2023.100216 . A. Sahu, P. K. Das, and S. Meher, “Recent advancements in machine learning and deep learning-based breast cancer detection using mammograms,” Phys. Medica, vol. 114, no. March, p. 103138, 2023, doi: 10.1016/j.ejmp.2023.103138 . S. Balasubramaniam, Y. Velmurugan, D. Jaganathan, and S. Dhanasekaran, “A Modified LeNet CNN for Breast Cancer Diagnosis in Ultrasound Images,” Diagnostics, vol. 13, no. 17, pp. 1–28, 2023, doi: 10.3390/diagnostics13172746 . E. Ahishakiye, M. B. Van Gijzen, J. Tumwiine, R. Wario, and J. Obungoloch, “A survey on deep learning in medical image reconstruction,” Intell. Med., vol. 1, no. 3, pp. 118–127, 2021, doi: 10.1016/j.imed.2021.03.003 . A. Khan, A. Sohail, U. Zahoora, and A. S. Qureshi, “A survey of the recent architectures of deep convolutional neural networks,” Artif. Intell. Rev., vol. 53, no. 8, pp. 5455–5516, 2020, doi: 10.1007/s10462-020-09825-6 . X. Han et al. , “Pre-trained models: Past, present and future,” AI Open , vol. 2, no. June 2021, pp. 225–250, 2021, doi: 10.1016/j.aiopen.2021.08.002 . D. Shah, M. Asmat, U. Khan, M. Abrar, and M. Tahir, “Optimizing Breast Cancer Detection With an Ensemble Deep Learning Approach,” Int. J. Intell. Syst. , vol. 2024, 2024, doi: 10.1155/2024/5564649 . T. A. Toma et al. , “Breast Cancer Detection Based on Simplified Deep Learning Technique With Histopathological Image Using BreaKHis Database,” Radio Sci., vol. 58, no. 11, 2023, doi: 10.1029/2023RS007761 . M. Wojciuk, Z. Swiderska-Chadaj, K. Siwek, and A. Gertych, “Improving classification accuracy of fine-tuned CNN models: Impact of hyperparameter optimization,” Heliyon, vol. 10, no. 5, p. e26586, 2024, doi: 10.1016/j.heliyon.2024.e26586 . J. Bergstra and Y. Bengio, “Random Search for Hyper-Parameter Optimization,” J. Mach. Learn., vol. 13, pp. 90–94, 2022, doi: 10.1145/3575882.3575900 . J. Wu, X. Y. Chen, H. Zhang, L. D. Xiong, H. Lei, and S. H. Deng, “Hyperparameter optimization for machine learning models based on Bayesian optimization,” J. Electron. Sci. Technol., vol. 17, no. 1, pp. 26–40, 2019, doi: 10.11989/JEST.1674-862X.80904120 . M. Alotaibi et al. , “Breast cancer classification based on convolutional neural network and image fusion approaches using ultrasound images,” Heliyon , vol. 9, no. 11, p. e22406, 2023, doi: 10.1016/j.heliyon.2023.e22406 . A. Golatkar, D. Anand, and A. Sethi, “Classification of Breast Cancer Histology Using Deep Learning,” Lect. Notes Comput. Sci. (including Subser. Lect. Notes Artif. Intell. Lect. Notes Bioinformatics), vol. 10882 LNCS, pp. 837–844, 2018, doi: 10.1007/978-3-319-93000-8_95 . Y. Shen et al. , “Leveraging Transformers to Improve Breast Cancer Classification and Risk Assessment with Multi-modal and Longitudinal Data,” 2023, [Online]. Available: http://arxiv.org/abs/2311.03217 E. Michael, H. Ma, H. Li, and S. Qi, “An Optimized Framework for Breast Cancer Classification Using Machine Learning,” Biomed Res. Int. , vol. 2022, 2022, doi: 10.1155/2022/8482022 . F. Shahidi, S. M. Daud, H. Abas, N. A. Ahmad, and N. Maarop, “Breast cancer classification using deep learning approaches and histopathology image: A comparison study,” IEEE Access, vol. 8, pp. 187531–187552, 2020, doi: 10.1109/ACCESS.2020.3029881 . M. Nasser and U. K. Yusof, “Deep Learning Based Methods for Breast Cancer Diagnosis: A Systematic Review and Future Direction,” Diagnostics, vol. 13, no. 1, 2023, doi: 10.3390/diagnostics13010161 . S. Zakareya, H. Izadkhah, and J. Karimpour, “A New Deep-Learning-Based Model for Breast Cancer Diagnosis from Medical Images,” Diagnostics, vol. 13, no. 11, pp. 1–23, 2023, doi: 10.3390/diagnostics13111944 . M. D. Ali et al. , “Breast Cancer Classification through Meta-Learning Ensemble Technique Using Convolution Neural Networks,” Diagnostics, vol. 13, no. 13, 2023, doi: 10.3390/diagnostics13132242 . T. Islam et al. , “Predictive modeling for breast cancer classification in the context of Bangladeshi patients by use of machine learning approach with explainable AI,” Sci. Rep., vol. 14, no. 1, pp. 1–17, 2024, doi: 10.1038/s41598-024-57740-5 . W. Lee, H. Lee, H. Lee, E. K. Park, H. Nam, and T. Kooi, “Transformer-based Deep Neural Network for Breast Cancer Classification on Digital Breast Tomosynthesis Images,” Radiol. Artif. Intell., vol. 5, no. 3, 2023, doi: 10.1148/ryai.220159 . G. L. Baroni, L. Rasotto, K. Roitero, A. Tulisso, C. Di Loreto, and V. Della Mea, “Optimizing Vision Transformers for Histopathology: Pretraining and Normalization in Breast Cancer Classification,” J. Imaging, vol. 10, no. 5, 2024, doi: 10.3390/jimaging10050108 . W. Al-Dhabyani, M. Gomaa, H. Khaled, and A. Fahmy, “Dataset of breast ultrasound images,” Data Br., vol. 28, p. 104863, 2020, doi: 10.1016/j.dib.2019.104863 . J. Li, “Assessing the accuracy of predictive models for numerical data: Not r nor r2, why not? Then what?,” PLoS One, vol. 12, no. 8, pp. 1–16, 2017, doi: 10.1371/journal.pone.0183250 . K. M. M. Uddin, N. Biswas, S. T. Rikta, and S. K. Dey, “Machine learning-based diagnosis of breast cancer utilizing feature optimization technique,” Comput. Methods Programs Biomed. Updat., vol. 3, no. February, p. 100098, 2023, doi: 10.1016/j.cmpbup.2023.100098 . Additional Declarations No competing interests reported. Cite Share Download PDF Status: Published Journal Publication published 29 May, 2025 Read the published version in Discover Artificial Intelligence → Version 1 posted Editorial decision: Revision requested 24 Feb, 2025 Reviews received at journal 22 Feb, 2025 Reviews received at journal 14 Feb, 2025 Reviewers agreed at journal 13 Feb, 2025 Reviewers agreed at journal 07 Feb, 2025 Reviews received at journal 07 Feb, 2025 Reviewers agreed at journal 07 Feb, 2025 Reviewers agreed at journal 07 Feb, 2025 Reviews received at journal 06 Feb, 2025 Reviewers agreed at journal 06 Feb, 2025 Reviewers agreed at journal 21 Nov, 2024 Reviewers agreed at journal 20 Nov, 2024 Reviewers agreed at journal 20 Nov, 2024 Reviewers invited by journal 20 Nov, 2024 Editor assigned by journal 11 Nov, 2024 Submission checks completed at journal 08 Nov, 2024 First submitted to journal 25 Oct, 2024 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-5333695","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":382309042,"identity":"a6243818-ebb9-4584-888e-86f770d681ea","order_by":0,"name":"Emmanuel Ahishakiye","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA1ElEQVRIiWNgGAWjYDACCcYGZhDNDyISCkjRItkA0mJAlBYGBrAWgwNgkggd/LObmz8XVGyTNz6/OvHDAwMGeX6xAwQsuXOwTXrGmduG22683SwBdJjhzNkJBKy5kdjGzNt2m3HbjbMbQFoSDG4T0CJ/I7H5M++/2/abZ5zd/IMoLQY3EhukeRtuJ27g791GnC2GQIdJ8xy7nTzjBu82iwQDCcJ+kbuR/vgzT81t2/7+s5tv/qiwkeeXJqAFASTAKiWIVQ4C/AdIUT0KRsEoGAUjCQAAen1JQlXcikQAAAAASUVORK5CYII=","orcid":"","institution":"Kyambogo University","correspondingAuthor":true,"prefix":"","firstName":"Emmanuel","middleName":"","lastName":"Ahishakiye","suffix":""},{"id":382309043,"identity":"5ac53b66-ee96-4ff5-b215-3ccd09c0ea31","order_by":1,"name":"Fredrick Kanobe","email":"","orcid":"","institution":"Kyambogo University","correspondingAuthor":false,"prefix":"","firstName":"Fredrick","middleName":"","lastName":"Kanobe","suffix":""}],"badges":[],"createdAt":"2024-10-25 16:23:11","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-5333695/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-5333695/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1007/s44163-025-00335-4","type":"published","date":"2025-05-29T15:57:10+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":71054763,"identity":"e5818462-87f3-43af-8e93-e00ebf70bb94","added_by":"auto","created_at":"2024-12-10 16:07:19","extension":"jpeg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":889065,"visible":true,"origin":"","legend":"\u003cp\u003eThe pseudo-code of the proposed model\u003c/p\u003e","description":"","filename":"floatimage1.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-5333695/v1/4774e1d037d6f60f21d33ebe.jpeg"},{"id":71056168,"identity":"668c649d-3eb5-43f2-a1ff-77d4e251d165","added_by":"auto","created_at":"2024-12-10 16:15:18","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":29420,"visible":true,"origin":"","legend":"\u003cp\u003eResults on the confusion matrix\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-5333695/v1/82b54b78a2f6ab9d7b51f376.png"},{"id":71054760,"identity":"ef724402-8460-4ef2-a4e1-e21d1f31d80f","added_by":"auto","created_at":"2024-12-10 16:07:18","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":56859,"visible":true,"origin":"","legend":"\u003cp\u003eResults of the AUC\u003c/p\u003e","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-5333695/v1/6bf7c838a2a9a607cfc76dea.png"},{"id":71054761,"identity":"00552056-a117-408b-be47-cb6cd4409eee","added_by":"auto","created_at":"2024-12-10 16:07:18","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":34778,"visible":true,"origin":"","legend":"\u003cp\u003eResults on Loss\u003c/p\u003e","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-5333695/v1/1be402697c6e98e7c3d5761a.png"},{"id":83782822,"identity":"c40ca701-a39f-4e08-9d70-e508b1153d7f","added_by":"auto","created_at":"2025-06-02 16:07:04","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1819284,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-5333695/v1/ffb325e5-95a6-46b3-832d-e21c4d318d8f.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Breast Cancer Classification Using Breast Ultrasound Images with a Hybrid of Transfer Learning and Bayesian-Optimized Fast Learning Network","fulltext":[{"header":"1. Background of the Study","content":"\u003cp\u003eBreast cancer is the most diagnosed cancer in women worldwide accounting for 11.7% of all new cases. An approximate number of 670,000 deaths due to the disease were recorded in 2022 alone [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. In 2022 alone, 2.3\u0026nbsp;million new cases of breast cancer were diagnosed, making this cancer type a big concern for public health and also the leading cause of death from cancer in females around the world. It is very important to diagnose cancer at an early stage, as it can remarkably improve the outcome of a patient. For instance, the five-year survival rate for localized and early-stage breast cancer is 99%, while it is only 28% when the diagnosed cases reach a farther stage onward [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eMammography, ultrasound, and MRI are the conventional medical imaging modalities employed in detecting and diagnosing breast cancer [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]. However, accurately interpreting these images can be challenging, even for experienced radiologists [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]. The sensitivity ranged from 25\u0026thinsp;\u0026minus;\u0026thinsp;58% for mammography, 33\u0026thinsp;\u0026minus;\u0026thinsp;52% for ultrasound, 48\u0026thinsp;\u0026minus;\u0026thinsp;67 for mammography plus ultrasound, and 71\u0026thinsp;\u0026minus;\u0026thinsp;100% for MRI in women with dense breast tissue, leading to missed detections and false negatives [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]. Therefore, automatic schemes are highly wanted to assist radiologists in correctly detecting the type of breast cancer. Medical image analysis has been greatly improved due to the advances in machine learning and deep learning methods [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e] [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]. These methods provide automated and objective support to classify breast cancer, with minimal or no dependence on manual feature extraction [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]. Among deep learning methods, CNN has been very successful in the classification tasks of breast cancer by automatically learning complex patterns from raw image data. For example, some CNN-based methods applied to breast ultrasound images have reached a very good performance in classifying positive versus negative cases [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]. However, deep learning models are confronted with such issues as requiring a lot of labeled data, running overfitting risks in the case of small datasets, and high computational loadings [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e] [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eOne of the various strategies that have been employed to resolve this is transfer learning [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]. Pre-trained models on large datasets like ImageNet have enabled knowledge acquired from general image classification tasks to be transferred for the analysis of medical images, thereby avoiding the need for labeled data in great quantity. Applications of transfer learning in breast cancer detection have been made using VGG16, ResNet, and DenseNet models and reached excellent classification accuracies [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]. Among them, DenseNet201 achieved the best performance in feature extraction due to its architecture wherein all the layers are connected densely to allow improved gradient flow and propagation of features [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]. While transfer learning enhances model performance, another critical challenge in practice is the optimization of hyperparameters of the learning model, such as the number of neurons, the learning rate, or the dropout rate [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]. Traditional methods, such as grid search and random search, are rather inefficient for the exploration of high-dimensional hyperparameter spaces of deep learning models [\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e]. Bayesian optimization, on the other hand, is a more efficient alternative that uses probabilistic models to guide the search for optimal hyperparameters [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]. This approach has outperformed the traditional methods in fine-tuning complex models. Bayesian optimization works particularly well for lightweight models such as the Fast Learning Network, which, if hyperparameters are carefully optimized, can give high accuracy with fast training speeds.\u003c/p\u003e \u003cp\u003eIntegration of transfer learning and Bayesian-optimized FLN is a very promising approach toward detection and classification in the case of breast cancer. This hybrid approach features extraction from breast ultrasound images, which uses a deep pretrained model like DenseNet201. These are then fed into an FLN model optimized using Bayesian techniques for finding the best hyperparameters. It was hypothesized that the proposed model would result in an improvement in the classification performance with limited data and model complexity by combining the strengths of transfer learning in effective feature extraction with Bayesian optimization in efficient model tuning within one hybrid framework. The proposed model is efficient and accurate for the classification of breast cancer and can potentially improve clinical decision-making and patient care. This study has been developed for SDG 3: \"\u003cem\u003eEnsure healthy lives and promote well-being for all at all ages\u003c/em\u003e,\" within the theme of artificial intelligence for development.\u003c/p\u003e"},{"header":"2. Literature Review","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003e2.1 Machine Learning in Breast Cancer Classification\u003c/h2\u003e \u003cp\u003eThe study [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e] proposed a three-step image processing approach that included RGB fusion, region of interest highlighting, and speckle noise filtering utilizing a block-matching three-dimensional filtering technique. This approach improves performance and broadens the generalization of deep learning models. Three datasets were employed in the study for transfer learning: BUSI (780 images), Dataset B (162 images), and KAIMRC (5693 images) using a deep learning model (VGG19). The model with the suggested preprocessing step outperformed the model without preprocessing for each dataset when tested using a fivefold cross-validation procedure on the BUSI and KAIMRC datasets. The deep learning classification model for breast cancer performs better with the proposed image processing method.\u003c/p\u003e \u003cp\u003eThe study (Uddin et al., 2023) employed the Wisconsin Breast Cancer Dataset (WBCD) as a training set from the UCI machine learning library to examine the effectiveness of the different machine learning techniques. To compare and analyze breast cancer into benign and malignant tumors, various machine learning classifiers have been used, including support vector machines (SVM), Random Forests (RF), K-nearest neighbors (K-NN), Decision Trees (DT), Na\u0026iuml;ve Bayes (NB), Logistic Regressions (LR), AdaBoost (AB), Gradient Boosting (GB), Multi-layer Perceptrons (MLP), Nearest Cluster Classifiers (NCC), and voting classifiers (VC). The research reveals that the voting classifier has the lowest mistake rate and the best accuracy, at 98.77%.\u003c/p\u003e \u003cp\u003eThe study [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e] proposed a deep learning-based technique that uses an Inception-v3 convolutional neural network to fine-tune images of breast tissue stained with Hematoxylin and Eosin (H\u0026amp;E). It is necessary to categorize these images into four groups: normal tissue, benign lesions, in situ cancer, and invasive carcinoma. Through majority vote over the nuclear classes, the class of the overall image is established. The findings showed a 93% accuracy rate for non-cancer (normal or benign) versus malignant (in situ or invasive carcinoma) and an average accuracy of 85% across the four classes.\u003c/p\u003e \u003cp\u003eThe study [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e] introduced a neural network called the Multi-modal Transformer (MMT), which uses ultrasound and mammography in concert to identify people with cancer and calculate the probability of developing cancer in those who are currently cancer-free. MMT uses self-attention to aggregate multimodal data and compares the present examinations to previous images to monitor changes in tissue over time. MMT surpasses robust uni-modal baselines with an AUROC of 0.943 in detecting pre-existing tumors after being trained on 1.3\u0026nbsp;million tests. With an AUROC of 0.826 for 5-year risk prediction, MMT outperforms earlier mammography-based risk models.\u003c/p\u003e \u003cp\u003eThe study [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e] proposed a computer-aided diagnosis (CAD) system that is capable of producing an optimal algorithm on its own. 13 of the 185 available features are used to train machine learning. Five machine-learning classifiers were employed to distinguish between benign and malignant tumors. Based on a machine learning classifier, the experimental findings demonstrated Bayesian optimization using a tree-structured Parzen estimator for 10-fold cross-validation. Outperforming the other four classifiers, the LightGBM classifier achieves 99.86% accuracy, 100.0% precision, 99.60% recall, and 99.80% for the FI score.\u003c/p\u003e \u003cp\u003eThe study [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e] examined the different Deep-learning models for categorizing photos of breast cancer histology. For the ImageNet database, models such as ResNeXt, Dual Path Net, SENet, and NASNet have been found to produce the most advanced results. The study also looked at the Inception-ResNet-V2 architecture, which produced better comparison results and produced the greatest results for binary and eight classifications.\u003c/p\u003e \u003cp\u003eThe study [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e] provided a thorough literature assessment of deep learning-based techniques for detecting breast cancer, which can help researchers and practitioners comprehend the difficulties and emerging trends in the field. With an emphasis on genomic and histopathological imaging data, various deep learning-based techniques for breast cancer identification are specifically examined. The review's findings showed that the Convolutional Neural Network (CNN) is the most widely used and accurate model for detecting breast cancer and that the most widely used technique for assessing performance is accuracy measures.\u003c/p\u003e \u003cp\u003eTo enhance the detection of breast cancer classification, the study [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e] proposed a new deep model to classify breast cancer. It was motivated by two cutting-edge deep networks, GoogLeNet, and residual block, and developed several new characteristics. The accuracy of the suggested model on ultrasound and breast histopathology images was 93% and 95%, respectively.\u003c/p\u003e \u003cp\u003eThe study [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e] proposed a Meta-Learning Ensemble Method for Breast Cancer Classification Using Convolution Neural Networks. Using a meta-learning framework, the proposed approach combined several CNN models, such as InceptionV3, ResNet50, and DenseNet121, to increase generalization and achieve a 90% accuracy rate, especially when identifying malignant tumors.\u003c/p\u003e \u003cp\u003eThe study [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e] used a primary dataset to assess and compare five distinct machine learning techniques' classification accuracy, precision, recall, and F1 scores. We employed five distinct supervised machine learning methods to attain the best outcomes on our dataset: logistic regression, naive Bayes, decision tree, random forest, and XGBoost. This study's final evaluation revealed that XGBoost had the best model accuracy, at 97%.\u003c/p\u003e \u003cp\u003eTo identify breast cancer on digital breast tomosynthesis (DBT) images, the study [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e] suggested a method for creating an effective deep neural network model that takes into account context from nearby image segments. The two 3D models outperformed the per-section baseline model in classification on the test set of 655 DBT trials. When compared with the single-DBT-section baseline, the suggested transformer-based model showed a significant increase in AUC (0.88 vs 0.91, P\u0026thinsp;=\u0026thinsp;.002), sensitivity (81.0% vs 87.7%, P\u0026thinsp;=\u0026thinsp;.006), and specificity (80.5% vs 86.4%, P\u0026thinsp;\u0026lt;\u0026thinsp;.001) at clinically important operating points. While showing comparable classification results, the transformer-based model only utilized 25% of the floating-point operations per second that the 3D convolution model did.\u003c/p\u003e \u003cp\u003eThe study [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e] presented a self-attention Vision Transformer model designed especially for histological image classification of breast cancer. Pre-training, dimension scaling, data augmentation, color normalization techniques, patch overlap, and patch size configurations are some of the training strategies and configurations that we analyze to assess their effects on the performance of histology picture categorization. An accuracy rate of 0.91, 0.74, and 0.92 on the BACH, BRACS, and AIDPATH datasets are attained by the model built from a transformer pretrained on ImageNet. Pre-training on big generic datasets such as ImageNet appears to be beneficial, according to the results.\u003c/p\u003e \u003c/div\u003e"},{"header":"3. Materials and Methods","content":"\u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003e3.1 Data Description\u003c/h2\u003e \u003cp\u003eDatasets [\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e] were collected at Baheya Hospital. Ultrasound (US) images are usually grayscale. The data was collected from women between the ages of 25 and 75 years old. The number of patients is 600 female patients. The dataset consists of 780 images with an average image size of 500x500 pixels. The US dataset is divided into three categories: benign, malignant, and normal. There were 1100 images gathered at the start and following the data preprocessing, only 780 images remained. LOGIQ Agile and LOGIQ E9 ultrasound systems are the tools utilized in the scanning procedure. Typically, these devices are employed in high-quality imaging for cardiac, vascular, and radiological applications. They generate images with a resolution of 1280 x 1024.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003e3.2 Data preprocessing\u003c/h2\u003e \u003cp\u003eData preprocessing was done on all the Pap smear images to maintain uniform quality and relevance for model training. This was done to improve the diversity of the dataset, data augmentation approaches can be employed, including rotation, flipping, and scaling. This increases the generalization capability of the models. The pixel values are finally normalized into a standard range from [0, 1], to ensure that the nature of the input provided to machine learning algorithms is always uniform. These preprocessing steps attempt to improve the quality of images and make the dataset suitable for the accurate classification of cervical cancer.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec7\" class=\"Section2\"\u003e \u003ch2\u003e3.3 The Proposed Model\u003c/h2\u003e \u003cp\u003eThe proposed model for the detection and classification of breast cancer combines the strength of DenseNet201 in feature extraction with a Bayesian-optimized Fast Learning Network (FLN) for classification. DenseNet201 serves as a robust feature extractor by transferring learned features to breast cancer detection. Thereafter, features are extracted and a light Fully Connected FLN is used for classification. The fully connected layers are optimized with Bayesian optimization to fine-tune the hyperparameters. The finetuned hyperparameters are learning rate, number of neurons, and dropouts. This further assists in the improvement of the model generalization capability with amplified performance. The model then yields class probabilities and can detect benign, malignant, and normal. The proposed model was evaluated using various performance metrics including accuracy, precision, recall, F1-score, and AUC. Besides that, confusion matrices and AUC curves were also employed to evaluate the performance of this model. The proposed approach of transfer learning along with Bayesian optimization aims at high accuracy with reliable classification in breast cancer detection.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003e3.4 The Pseudocode of the Proposed Model\u003c/h2\u003e \u003cp\u003eThe Pseudocode describes the model, which involves the following steps: a. Pre-processing of images of breast cancers; b. Loading the DenseNet201 pre-trained feature extractor, from which it removes the last classification layers. It is followed by a Fast Learning Network - FLN, containing two fully connected layers optimized with Bayesian optimization in tuning hyper-parameters such as learning rate and dropout. These class probabilities, benign, malignant, and normal are output through a softmax layer at the end of the FLN. Then we compile the model using the Adam optimizer and categorical cross-entropy loss. We test the performance of the model on accuracy, F1-score, precision, recall, and AUC together with the confusion matrix and AUC curves.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec9\" class=\"Section2\"\u003e \u003ch2\u003e3.5 Model Evaluation\u003c/h2\u003e \u003cp\u003eThe proposed models were assessed using several metrics: accuracy, cross-entropy loss, precision, recall, F1-score, and loss function. Accuracy measures the ratio of correctly predicted outcomes to the total number of predictions [\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e]. It was chosen for this task because it provides a quick assessment of a model's performance and is particularly effective for binary classification problems. Also, the cross-entropy metric measures the model's error and dissimilarity between predicted and actual values. The performance metrics are calculated using the formulas below:\u003cdiv id=\"Equa\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equa\" name=\"EquationSource\"\u003e\n$$\\:Accuracy=\\:\\frac{TP+TN}{TP+TN+FP+FN}\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\left(1\\right)$$\u003c/div\u003e\u003c/div\u003e\u003cdiv id=\"Equb\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equb\" name=\"EquationSource\"\u003e\n$$\\:F1-Score=\\:\\frac{2TP}{2TP+FP+FN}\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\left(2\\right)$$\u003c/div\u003e\u003c/div\u003e\u003cdiv id=\"Equc\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equc\" name=\"EquationSource\"\u003e\n$$\\:Precision=\\:\\frac{TP}{TP+FP}\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\left(3\\right)$$\u003c/div\u003e\u003c/div\u003e\u003cdiv id=\"Equd\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equd\" name=\"EquationSource\"\u003e\n$$\\:Recall,\\:Sensitivity=\\:\\frac{TP}{TP+FN}\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\left(4\\right)$$\u003c/div\u003e\u003c/div\u003e\u003cdiv id=\"Eque\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Eque\" name=\"EquationSource\"\u003e\n$$\\:cross\\:entropy\\:loss,\\:H\\left(y,\\:\\stackrel{\u0026acute;}{y}\\right)=-\\sum\\:_{i=1}^{n}{y}_{i}\\text{l}\\text{o}\\text{g}(\\stackrel{\u0026acute;}{{y}_{i}}\\left)\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\right(5)$$\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003ewhere \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:TP,\\:TN,\\:FP,\\:and\\:FN\\)\u003c/span\u003e\u003c/span\u003e stand for true positive, true negative, false positive, and false negative, respectively. Also, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:y\\)\u003c/span\u003e\u003c/span\u003e is the true probability distribution (the one-hot encoded true labels), \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\stackrel{\u0026acute;}{y}\\)\u003c/span\u003e\u003c/span\u003e is the predicted probability distribution (the model\u0026rsquo;s predicted probabilities for each class), \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:n\\)\u003c/span\u003e\u003c/span\u003e is the number of classes in the classification problem, and \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:log\\)\u003c/span\u003e\u003c/span\u003e is the natural logarithm function.\u003c/p\u003e \u003cp\u003eThe proposed model was implemented using Python 3.12.4, TensorFlow 2.16.0, Keras 3.4.0, scikit-learn (sklearn) 1.5.0, NumPy 2.0, Pandas 2.2.2 and Matplotlib 3.9.1. Also, a Windows 11 laptop, Intel Core I7 processor of 2.50 GHz, and 16GB RAM with 2GB of NVIDIA GeForce MX150 graphical processing unit (GPU) were used.\u003c/p\u003e \u003c/div\u003e"},{"header":"4. Results","content":"\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003e4.1 Comparative Performance of the Pre-Trained Transfer Learning Models\u003c/h2\u003e \u003cp\u003eTo determine the best pre-trained transfer learning model to integrate with the optimized FLN, we did a performance evaluation of some of the existing transfer learning models. In this work, Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e displays a few of the pre-trained transfer learning models that have been refined on the dataset used for cervical cancer. Because these models have already been trained, they can take advantage of feature representations that have been acquired from a wide range of different and sizable datasets. The last layers of each model were changed, and the cervical cancer dataset was used for additional retraining so that the learned characteristics of the model could adjust to this particular classification job. Changing the model parameters, particularly the upper layers, while maintaining fixed or low learning rates for the earlier layers that identify more general features like edges and textures is known as fine-tuning. Each model underwent five-fold cross-validation to evaluate its generalization on unknown data, and the results were recorded for training accuracy, validation accuracy, training loss, and validation loss. By fine-tuning, the models were able to focus on detecting features that are important for breast cancer diagnosis. DenseNet201 was chosen as the study's preferred model since it was the overall best in performance.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003ePerformance of the Pre-Trained Models\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eModel\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTraining Accuracy (%)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eValidation Accuracy (%)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eTraining Loss\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eValidation loss\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eVGG16\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e95.54\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e92.75\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.82\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e1.97\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eVGG19\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e95.36\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e91.14\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.89\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e1.95\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eResNet50\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e95.70\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e94.35\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.66\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e2.01\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eInceptionV3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e97.50\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e95.62\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.45\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e1.99\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDenseNet201\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e97.93\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e96.54\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.31\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e1.83\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMobileNetV3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e97.15\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e95.62\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.22\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.24\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003e4.2 Bayesian Optimization of Hyperparameters\u003c/h2\u003e \u003cp\u003eResults for the optimal values of the various tunings regarding FLN are highlighted in Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e, conducted through Bayesian optimization. In this model, the number of 471 hidden neurons is a reasonable balance between learning capability and model complexity. The learning rate is 0.000506, way below 1.0, which allows gradual weight updates, not overshooting, and makes the training process stable for more accuracy. The 44.18% dropout is a technique that reduces overfitting by randomly ignoring neurons during training to enhance the generalization capability of the model. Finally, Adam has been chosen as an optimizer because, with it, adaptive learning rate and momentum are possible, which allows the model performance to converge faster.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eBest Hyperparameters\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"2\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eParameter\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003evalue\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHidden Neurons\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e471\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLearning Rate\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.0005064479220813515\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDropout Rate\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.44180424373490723\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eOptimizer\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAdam\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003e4.3 Performance on Accuracy, F1 score, precision, and Recall\u003c/h2\u003e \u003cp\u003eThe proposed model shows very good performance compared to the models in the literature as shown in Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e. It has high accuracy of 96.79%, F1 score of 94.71%, precision of 96.81%, and recall of 93.48%. Thus, the model appears competitive since, besides the advantage in accuracy for several models like those by [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e] and [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e], it gives a more balanced set of metrics. Although this does not quite match the near-perfect performance seen in [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e], with an accuracy of 99.86% and perfect precision, the proposed model continues to show very good precision and recall, hence reliable and effective classification of cases of breast cancer. precision of 96.81% indeed depicts a strong ability to minimize false positives, hence making it a robust choice; however, there is still room for further optimization toward the highest-performing models.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003ePerformance of the model on performance metrics\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eModel\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAccuracy (%)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eF1 score (%)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003ePrecision (%)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eRecall (%)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e[\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e99.86\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e99.80\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e100.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e99.60\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e[\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e98.77\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e[\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e85.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e[\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e95.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e[\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e90.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e90.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e94.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e95.00\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e[\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e74.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eProposed\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e96.79\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e94.71\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e96.81\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e93.48\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003e4.4 Performance on Confusion Matrix\u003c/h2\u003e \u003cp\u003eThe results of the confusion matrix proposed model are shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e. The proposed model had a true positive of 165 for Benign and false positive of 11 and a false negative of 7, then a true positive of 72 for malignant with a false negative of 14 and a false positive of 7, and finally a true positive of 39 for normal cases with the false negative of 2 and false positive of 1. The model performs very well for the benign and normal classes, with a high number of true positives and relatively low false positives and negatives.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec15\" class=\"Section2\"\u003e \u003ch2\u003e4.4 Performance on AUC for each Class\u003c/h2\u003e \u003cp\u003eThe AUC is a measure of the performance of the ability of the model in differentiating classes. AUC ranges between 0 and 1, with values close to 1 indicating better performance. The proposed model obtained the values of AUC for classes as benign (0.96), malignant (0.95), and normal (0.98) demonstrating that the model did a great a great job in classifying all three classes of breast cancer with the most excellent ability to correctly identify the normal cases. The relatively lower AUC for the malignant class may be indicative of potential improvements but overall, the model demonstrated great performance. The AUC of the proposed model is shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec16\" class=\"Section2\"\u003e \u003ch2\u003e4.5 Results on Cross-Entropy Loss\u003c/h2\u003e \u003cp\u003eThe Cross-entropy loss quantifies how distant the predicted probability distribution is from the true distribution of target classes. A lower cross-entropy loss indicates that predictions are closer to real class labels. When it has a value near zero, that will mean very good accuracy; with higher values, the accuracy will be bad. During training time, a decreasing cross-entropy loss is an indication of a well-learning model that slowly gets better in its predictions. Figure\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e shows that with the increase in iteration, cross-entropy loss decreased both in training and validation, proving our proposed model's effectiveness.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"5. Discussion of the Results","content":"\u003cp\u003eThe proposed model for the classification of breast cancer was based on DenseNet201 as the base feature extractor with a Bayesian-Optimized Fast Learning Network. Overall, its performance was excellent in all metrics. The quantitative comparison of the results of the proposed method with the state-of-the-art pre-trained models is shown below in Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e; the best results corresponded to DenseNet201 with a validation accuracy of 96.54%, while in other implemented models, VGG16 had a high performance of 92.75% and the closest, ResNet50 achieved 94.35%. It also means that DenseNet201 has layered connectivity and uses its parameters effectively to emphasize the relevant features for diagnosing breast cancer; it is therefore well-motivated to be integrated with FLN. The fine-tuning enabled the model to adapt the pre-trained feature representations to the specifics of the cervical cancer dataset, thus enabling improvement in both classification accuracy and generalization.\u003c/p\u003e \u003cp\u003eThe optimized FLN\u0026rsquo;s hyperparameters, as presented in Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e, were crucial in further refining the model\u0026rsquo;s performance. The Bayesian optimization yielded a learning rate of 0.000506 and a dropout rate of 44.18%, with 471 hidden neurons to strike a balance between learning ability and model complexity. The chosen Adam optimizer enabled efficient convergence by adjusting learning rates adaptively, which facilitated the model\u0026rsquo;s ability to stabilize quickly while minimizing overfitting. These adjustments are reflected in the proposed model\u0026rsquo;s high accuracy of 96.79% and a strong F1 score of 94.71%, indicating effective handling of both false positives and false negatives, making it a robust solution for breast cancer classification. Hence, this proposed model is competitive with the literature's existing models. Although not near-perfect performance, as seen by [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e], with 99.86% accuracy and perfect precision, the proposed model still outperforms models like [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e] and [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e] in terms of both accuracy and precision. The proposed model had a precision of 96.81%, showing good performance in terms of reducing false positives, a quite important ability within a medical diagnosis environment while its recall of 93.48% shows very reliable sensitivity in depicting the actual positive cases.\u003c/p\u003e \u003cp\u003eThe robustness of the proposed model is further confirmed by the confusion matrix results, represented in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e, showing the effectiveness of the model, especially in the case of benign and normal cases. In this respect, with true positives of 165 for benign and 39 for normal, while having relatively low false positives and false negatives, the model has shown high sensitivity and specificity for these classes. While the malignant class has relatively somewhat higher misclassification rates, it shows 14 false negatives. This means additional tuning will be required to boost recall for malignant instances, by either augmenting training data on malignant features or using class-specific weighting during training itself. More so, from the perspective of AUC scores, the proposed model has the best class discrimination ability, with AUC values of 0.96 for benign, 0.95 for malignant, and 0.98 for normal. A high AUC in a normal class means outstanding ability in correctly identifying normal cases, whereas a little lower AUC value observed in malignant, 0.95, indicates that there may be some difficulties in distinguishing the malignant case from the rest. The strong overall performance in AUC reinforces the robustness of the model, but its performance could be further improved with targeted improvements in malignant classification.\u003c/p\u003e \u003cp\u003eThe results of cross-entropy loss prove that there has been effective learning throughout the training, clearly seen by the drop in both training and validation loss at each iteration. This downward slope of cross-entropy loss confirms that with time increasing, the model was aligning its predictions toward true labels by gradually increasing the accuracy. Fairly low cross-entropy loss at the end of training pinpoints the model's reliability and hence supports the conclusion that the proposed approach will be accurate and stable.\u003c/p\u003e \u003cp\u003eTherefore, the performance of the proposed model for the breast cancer classification is quite rounded and effectively balanced between accuracy, precision, and recall. Feature extraction through DenseNet201 coupled with optimized FLN thus performs reliable classification among benign, malignant, and normal classes. Although this model performs very well, especially in reducing false positives, the enhancement in sensitivity concerning malignant cases will easily bring it closer to the model performance of the top results reported in the literature.\u003c/p\u003e"},{"header":"6. Conclusion","content":"\u003cp\u003eThis paper proposes a hybrid model that involves DenseNet201 and a Bayesian optimization-based Fast Learning Network for efficient detection and classification of BC. The results showed that the proposed model leveraged pre-trained feature extraction afforded by DenseNet201, augmented by flexibility for fine-tuning by Bayesian-optimized FLN, achieved robust results on all metrics of accuracy at 96.79%, F1 score at 94.71%, precision at 96.81%, and recall at 93.48%. Besides, the high values of AUC for benign, malignant, and normal classes prove this model is reliable in classifying categories of interest in breast cancer. Although such results show competitiveness compared with state-of-the-art models, slight improvements might boost sensitivity, especially in the detection of malignancy. This concept has shown huge possibility in the combined application of transfer learning with Bayesian optimization toward improving the accuracy of classification in medical imaging for timely diagnosis in the clinical setting. Future studies may consider improvements in data augmentation and ensemble learning to increase model sensitivity and overall accuracy.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eAuthor contributions:\u0026nbsp;\u003c/strong\u003eEmmanuel Ahishakiye (EA) and Fredrick Kanobe (FK) designed the study. EA wrote the manuscript. FK edited and extensively reviewed the manuscript. All the authors approved the manuscript.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding:\u0026nbsp;\u003c/strong\u003eWe acknowledge the financial support from Kyambogo University's competitive grants.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData Availability:\u0026nbsp;\u003c/strong\u003eThe data utilized in this study is sourced from the Dataset of breast ultrasound images,” \u003cem\u003eData Br.\u003c/em\u003e, vol. 28, p. 104863, 2020\" which is publicly accessible at the following URL: https://www.kaggle.com/datasets/aryashah2k/breast-ultrasound-images-dataset. Accessed 10-October-2024. doi: 10.1016/j.dib.2019.104863.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCode availability:\u0026nbsp;\u003c/strong\u003eNot applicable.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare that they have no competing interests.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eWorld Health Organization, \u0026ldquo;Breast cancer.\u0026rdquo; [Online]. Available: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.who.int/news-room/fact-sheets/detail/breast-cancer\u003c/span\u003e\u003cspan address=\"https://www.who.int/news-room/fact-sheets/detail/breast-cancer\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAmerican Cancer Society, \u0026ldquo;Breast Cancer Facts \u0026amp; Figures.\u0026rdquo; Accessed: Oct. 25, 2024. [Online]. Available: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.cancer.org/research/cancer-facts-statistics/breast-cancer-facts-figures.html\u003c/span\u003e\u003cspan address=\"https://www.cancer.org/research/cancer-facts-statistics/breast-cancer-facts-figures.html\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eE. Abu Abeelh and Z. AbuAbeileh, \u0026ldquo;Comparative Effectiveness of Mammography, Ultrasound, and MRI in the Detection of Breast Carcinoma in Dense Breast Tissue: A Systematic Review,\u0026rdquo; Cureus, vol. 16, no. 4, 2024, doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.7759/cureus.59054\u003c/span\u003e\u003cspan address=\"10.7759/cureus.59054\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eK. Loizidou, R. Elia, and C. Pitris, \u0026ldquo;Computer-aided breast cancer detection and classification in mammography: A comprehensive review,\u0026rdquo; \u003cem\u003eComput. Biol. Med.\u003c/em\u003e, vol. 153, no. November 2022, p. 106554, 2023, doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.compbiomed.2023.106554\u003c/span\u003e\u003cspan address=\"10.1016/j.compbiomed.2023.106554\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eF. Sardanelli, V. Magni, G. Rossini, F. Kilburn-Toppin, N. A. Healy, and F. J. Gilbert, \u0026ldquo;The paradox of MRI for breast cancer screening: high-risk and dense breasts-available evidence and current practice,\u0026rdquo; Insights Imaging, vol. 15, no. 1, 2024, doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s13244-024-01653-4\u003c/span\u003e\u003cspan address=\"10.1186/s13244-024-01653-4\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eM. Li, Y. Jiang, Y. Zhang, and H. Zhu, \u0026ldquo;Medical image analysis using deep learning algorithms,\u0026rdquo; \u003cem\u003eFront. Public Heal.\u003c/em\u003e, vol. 11, no. November, pp. 1\u0026ndash;28, 2023, doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3389/fpubh.2023.1273253\u003c/span\u003e\u003cspan address=\"10.3389/fpubh.2023.1273253\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eP. K. Mall \u003cem\u003eet al.\u003c/em\u003e, \u0026ldquo;A comprehensive review of deep neural networks for medical image processing: Recent developments and future opportunities,\u0026rdquo; Healthc. Anal., vol. 4, no. June, p. 100216, 2023, doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.health.2023.100216\u003c/span\u003e\u003cspan address=\"10.1016/j.health.2023.100216\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eA. Sahu, P. K. Das, and S. Meher, \u0026ldquo;Recent advancements in machine learning and deep learning-based breast cancer detection using mammograms,\u0026rdquo; Phys. Medica, vol. 114, no. March, p. 103138, 2023, doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.ejmp.2023.103138\u003c/span\u003e\u003cspan address=\"10.1016/j.ejmp.2023.103138\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eS. Balasubramaniam, Y. Velmurugan, D. Jaganathan, and S. Dhanasekaran, \u0026ldquo;A Modified LeNet CNN for Breast Cancer Diagnosis in Ultrasound Images,\u0026rdquo; Diagnostics, vol. 13, no. 17, pp. 1\u0026ndash;28, 2023, doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3390/diagnostics13172746\u003c/span\u003e\u003cspan address=\"10.3390/diagnostics13172746\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eE. Ahishakiye, M. B. Van Gijzen, J. Tumwiine, R. Wario, and J. Obungoloch, \u0026ldquo;A survey on deep learning in medical image reconstruction,\u0026rdquo; Intell. Med., vol. 1, no. 3, pp. 118\u0026ndash;127, 2021, doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.imed.2021.03.003\u003c/span\u003e\u003cspan address=\"10.1016/j.imed.2021.03.003\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eA. Khan, A. Sohail, U. Zahoora, and A. S. Qureshi, \u0026ldquo;A survey of the recent architectures of deep convolutional neural networks,\u0026rdquo; Artif. Intell. Rev., vol. 53, no. 8, pp. 5455\u0026ndash;5516, 2020, doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s10462-020-09825-6\u003c/span\u003e\u003cspan address=\"10.1007/s10462-020-09825-6\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eX. Han \u003cem\u003eet al.\u003c/em\u003e, \u0026ldquo;Pre-trained models: Past, present and future,\u0026rdquo; \u003cem\u003eAI Open\u003c/em\u003e, vol. 2, no. June 2021, pp. 225\u0026ndash;250, 2021, doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.aiopen.2021.08.002\u003c/span\u003e\u003cspan address=\"10.1016/j.aiopen.2021.08.002\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eD. Shah, M. Asmat, U. Khan, M. Abrar, and M. Tahir, \u0026ldquo;Optimizing Breast Cancer Detection With an Ensemble Deep Learning Approach,\u0026rdquo; \u003cem\u003eInt. J. Intell. Syst.\u003c/em\u003e, vol. 2024, 2024, doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1155/2024/5564649\u003c/span\u003e\u003cspan address=\"10.1155/2024/5564649\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eT. A. Toma \u003cem\u003eet al.\u003c/em\u003e, \u0026ldquo;Breast Cancer Detection Based on Simplified Deep Learning Technique With Histopathological Image Using BreaKHis Database,\u0026rdquo; Radio Sci., vol. 58, no. 11, 2023, doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1029/2023RS007761\u003c/span\u003e\u003cspan address=\"10.1029/2023RS007761\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eM. Wojciuk, Z. Swiderska-Chadaj, K. Siwek, and A. Gertych, \u0026ldquo;Improving classification accuracy of fine-tuned CNN models: Impact of hyperparameter optimization,\u0026rdquo; Heliyon, vol. 10, no. 5, p. e26586, 2024, doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.heliyon.2024.e26586\u003c/span\u003e\u003cspan address=\"10.1016/j.heliyon.2024.e26586\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJ. Bergstra and Y. Bengio, \u0026ldquo;Random Search for Hyper-Parameter Optimization,\u0026rdquo; J. Mach. Learn., vol. 13, pp. 90\u0026ndash;94, 2022, doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1145/3575882.3575900\u003c/span\u003e\u003cspan address=\"10.1145/3575882.3575900\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJ. Wu, X. Y. Chen, H. Zhang, L. D. Xiong, H. Lei, and S. H. Deng, \u0026ldquo;Hyperparameter optimization for machine learning models based on Bayesian optimization,\u0026rdquo; J. Electron. Sci. Technol., vol. 17, no. 1, pp. 26\u0026ndash;40, 2019, doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.11989/JEST.1674-862X.80904120\u003c/span\u003e\u003cspan address=\"10.11989/JEST.1674-862X.80904120\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eM. Alotaibi \u003cem\u003eet al.\u003c/em\u003e, \u0026ldquo;Breast cancer classification based on convolutional neural network and image fusion approaches using ultrasound images,\u0026rdquo; \u003cem\u003eHeliyon\u003c/em\u003e, vol. 9, no. 11, p. e22406, 2023, doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.heliyon.2023.e22406\u003c/span\u003e\u003cspan address=\"10.1016/j.heliyon.2023.e22406\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eA. Golatkar, D. Anand, and A. Sethi, \u0026ldquo;Classification of Breast Cancer Histology Using Deep Learning,\u0026rdquo; Lect. Notes Comput. Sci. (including Subser. Lect. Notes Artif. Intell. Lect. Notes Bioinformatics), vol. 10882 LNCS, pp. 837\u0026ndash;844, 2018, doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/978-3-319-93000-8_95\u003c/span\u003e\u003cspan address=\"10.1007/978-3-319-93000-8_95\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eY. Shen \u003cem\u003eet al.\u003c/em\u003e, \u0026ldquo;Leveraging Transformers to Improve Breast Cancer Classification and Risk Assessment with Multi-modal and Longitudinal Data,\u0026rdquo; 2023, [Online]. Available: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://arxiv.org/abs/2311.03217\u003c/span\u003e\u003cspan address=\"http://arxiv.org/abs/2311.03217\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eE. Michael, H. Ma, H. Li, and S. Qi, \u0026ldquo;An Optimized Framework for Breast Cancer Classification Using Machine Learning,\u0026rdquo; \u003cem\u003eBiomed Res. Int.\u003c/em\u003e, vol. 2022, 2022, doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1155/2022/8482022\u003c/span\u003e\u003cspan address=\"10.1155/2022/8482022\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eF. Shahidi, S. M. Daud, H. Abas, N. A. Ahmad, and N. Maarop, \u0026ldquo;Breast cancer classification using deep learning approaches and histopathology image: A comparison study,\u0026rdquo; IEEE Access, vol. 8, pp. 187531\u0026ndash;187552, 2020, doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1109/ACCESS.2020.3029881\u003c/span\u003e\u003cspan address=\"10.1109/ACCESS.2020.3029881\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eM. Nasser and U. K. Yusof, \u0026ldquo;Deep Learning Based Methods for Breast Cancer Diagnosis: A Systematic Review and Future Direction,\u0026rdquo; Diagnostics, vol. 13, no. 1, 2023, doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3390/diagnostics13010161\u003c/span\u003e\u003cspan address=\"10.3390/diagnostics13010161\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eS. Zakareya, H. Izadkhah, and J. Karimpour, \u0026ldquo;A New Deep-Learning-Based Model for Breast Cancer Diagnosis from Medical Images,\u0026rdquo; Diagnostics, vol. 13, no. 11, pp. 1\u0026ndash;23, 2023, doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3390/diagnostics13111944\u003c/span\u003e\u003cspan address=\"10.3390/diagnostics13111944\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eM. D. Ali \u003cem\u003eet al.\u003c/em\u003e, \u0026ldquo;Breast Cancer Classification through Meta-Learning Ensemble Technique Using Convolution Neural Networks,\u0026rdquo; Diagnostics, vol. 13, no. 13, 2023, doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3390/diagnostics13132242\u003c/span\u003e\u003cspan address=\"10.3390/diagnostics13132242\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eT. Islam \u003cem\u003eet al.\u003c/em\u003e, \u0026ldquo;Predictive modeling for breast cancer classification in the context of Bangladeshi patients by use of machine learning approach with explainable AI,\u0026rdquo; Sci. Rep., vol. 14, no. 1, pp. 1\u0026ndash;17, 2024, doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/s41598-024-57740-5\u003c/span\u003e\u003cspan address=\"10.1038/s41598-024-57740-5\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eW. Lee, H. Lee, H. Lee, E. K. Park, H. Nam, and T. Kooi, \u0026ldquo;Transformer-based Deep Neural Network for Breast Cancer Classification on Digital Breast Tomosynthesis Images,\u0026rdquo; Radiol. Artif. Intell., vol. 5, no. 3, 2023, doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1148/ryai.220159\u003c/span\u003e\u003cspan address=\"10.1148/ryai.220159\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eG. L. Baroni, L. Rasotto, K. Roitero, A. Tulisso, C. Di Loreto, and V. Della Mea, \u0026ldquo;Optimizing Vision Transformers for Histopathology: Pretraining and Normalization in Breast Cancer Classification,\u0026rdquo; J. Imaging, vol. 10, no. 5, 2024, doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3390/jimaging10050108\u003c/span\u003e\u003cspan address=\"10.3390/jimaging10050108\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eW. Al-Dhabyani, M. Gomaa, H. Khaled, and A. Fahmy, \u0026ldquo;Dataset of breast ultrasound images,\u0026rdquo; Data Br., vol. 28, p. 104863, 2020, doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.dib.2019.104863\u003c/span\u003e\u003cspan address=\"10.1016/j.dib.2019.104863\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJ. Li, \u0026ldquo;Assessing the accuracy of predictive models for numerical data: Not r nor r2, why not? Then what?,\u0026rdquo; PLoS One, vol. 12, no. 8, pp. 1\u0026ndash;16, 2017, doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1371/journal.pone.0183250\u003c/span\u003e\u003cspan address=\"10.1371/journal.pone.0183250\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eK. M. M. Uddin, N. Biswas, S. T. Rikta, and S. K. Dey, \u0026ldquo;Machine learning-based diagnosis of breast cancer utilizing feature optimization technique,\u0026rdquo; Comput. Methods Programs Biomed. Updat., vol. 3, no. February, p. 100098, 2023, doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.cmpbup.2023.100098\u003c/span\u003e\u003cspan address=\"10.1016/j.cmpbup.2023.100098\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"discover-artificial-intelligence","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"diai","sideBox":"Learn more about [Discover Artificial Intelligence](https://www.springer.com/44163)","snPcode":"","submissionUrl":"","title":"Discover Artificial Intelligence","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Discover Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Breast Cancer, Machine learning, Deep learning, Fast Learning Network, Bayesian Optimization","lastPublishedDoi":"10.21203/rs.3.rs-5333695/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-5333695/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground\u003c/h2\u003e \u003cp\u003eBreast cancer remains the most frequent cancer diagnosed in females, resulting in high mortality rates worldwide. Approximately 2.3\u0026nbsp;million cases are diagnosed annually. If it is detected at an early stage, the rate of survival is significantly improved; therefore, there is an urgent need for techniques that can be used for its effective diagnosis.\u003c/p\u003e\u003ch2\u003eMethod\u003c/h2\u003e \u003cp\u003eThe study aimed to present a hybrid model for breast cancer classification by employing DenseNet201 as a feature extractor and Bayesian-Optimized Fast Learning Network (FLN) as a classifier. The pre-trained DenseNet201 extracts high-quality features from breast ultrasound images on large datasets, which get classified through an FLN optimized using Bayesian techniques for hyperparameter tuning.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003eThe model performed well by achieving an accuracy of 96.79%, 94.71% F1 score, 96.81% precision, and 93.48% recall, while the AUC for benign, malignant, and normal cases was found to be 0.96, 0.95, and 0.98, respectively. Cross-entropy loss metrics further validated the model on its robust training and validation.\u003c/p\u003e\u003ch2\u003eConclusion\u003c/h2\u003e \u003cp\u003eThere is a great potential that this proposed model could enhance breast cancer diagnosis. This indeed is a reliable and efficient clinical solution for application.\u003c/p\u003e","manuscriptTitle":"Breast Cancer Classification Using Breast Ultrasound Images with a Hybrid of Transfer Learning and Bayesian-Optimized Fast Learning Network","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-12-10 16:07:14","doi":"10.21203/rs.3.rs-5333695/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2025-02-24T10:12:24+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-02-22T06:25:39+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-02-14T12:23:27+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"248265885419096431745626281322199517674","date":"2025-02-14T04:45:57+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"89963037321654279801721925715839992069","date":"2025-02-08T04:07:49+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-02-07T09:52:24+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"256281750665312676051232032746188626675","date":"2025-02-07T09:32:21+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"186849457887145999507063616719317807513","date":"2025-02-07T08:54:15+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-02-07T04:06:57+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"96892821471936564047941797305231555772","date":"2025-02-06T05:52:16+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"234372448625706911437432938523048641425","date":"2024-11-21T13:49:44+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"129993123954502737735809267880063530834","date":"2024-11-20T08:22:31+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"179099077183009442065055447810054543374","date":"2024-11-20T07:31:12+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2024-11-20T07:24:24+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2024-11-11T16:40:26+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2024-11-08T12:37:37+00:00","index":"","fulltext":""},{"type":"submitted","content":"Discover Artificial Intelligence","date":"2024-10-25T16:12:18+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"discover-artificial-intelligence","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"diai","sideBox":"Learn more about [Discover Artificial Intelligence](https://www.springer.com/44163)","snPcode":"","submissionUrl":"","title":"Discover Artificial Intelligence","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Discover Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"01c87f19-b1db-4f06-a627-9832cad49d80","owner":[],"postedDate":"December 10th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[],"tags":[],"updatedAt":"2025-06-02T15:59:50+00:00","versionOfRecord":{"articleIdentity":"rs-5333695","link":"https://doi.org/10.1007/s44163-025-00335-4","journal":{"identity":"discover-artificial-intelligence","isVorOnly":false,"title":"Discover Artificial Intelligence"},"publishedOn":"2025-05-29 15:57:10","publishedOnDateReadable":"May 29th, 2025"},"versionCreatedAt":"2024-12-10 16:07:14","video":"","vorDoi":"10.1007/s44163-025-00335-4","vorDoiUrl":"https://doi.org/10.1007/s44163-025-00335-4","workflowStages":[]},"version":"v1","identity":"rs-5333695","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-5333695","identity":"rs-5333695","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}