Efficient Chest X-Ray Feature Extraction and Feature Fusion for Pneumonia Detection Using Lightweight Pretrained Deep Learning Models

Efficient Chest X-Ray Feature Extraction and Feature Fusion for Pneumonia Detection Using Lightweight Pretrained Deep Learning Models | medRxiv /* */ /* */ <!-- <!-- /*! * yepnope1.5.4 * (c) WTFPL, GPLv2 */ (function(a,b,c){function d(a){return"[object Function]"==o.call(a)}function e(a){return"string"==typeof a}function f(){}function g(a){return!a||"loaded"==a||"complete"==a||"uninitialized"==a}function h(){var a=p.shift();q=1,a?a.t?m(function(){("c"==a.t?B.injectCss:B.injectJs)(a.s,0,a.a,a.x,a.e,1)},0):(a(),h()):q=0}function i(a,c,d,e,f,i,j){function k(b){if(!o&&g(l.readyState)&&(u.r=o=1,!q&&h(),l.onload=l.onreadystatechange=null,b)){"img"!=a&&m(function(){t.removeChild(l)},50);for(var d in y[c])y[c].hasOwnProperty(d)&&y[c][d].onload()}}var j=j||B.errorTimeout,l=b.createElement(a),o=0,r=0,u={t:d,s:c,e:f,a:i,x:j};1===y[c]&&(r=1,y[c]=[]),"object"==a?l.data=c:(l.src=c,l.type=a),l.width=l.height="0",l.onerror=l.onload=l.onreadystatechange=function(){k.call(this,r)},p.splice(e,0,u),"img"!=a&&(r||2===y[c]?(t.insertBefore(l,s?null:n),m(k,j)):y[c].push(l))}function j(a,b,c,d,f){return q=0,b=b||"j",e(a)?i("c"==b?v:u,a,b,this.i++,c,d,f):(p.splice(this.i++,0,a),1==p.length&&h()),this}function k(){var a=B;return a.loader={load:j,i:0},a}var l=b.documentElement,m=a.setTimeout,n=b.getElementsByTagName("script")[0],o={}.toString,p=[],q=0,r="MozAppearance"in l.style,s=r&&!!b.createRange().compareNode,t=s?l:n.parentNode,l=a.opera&&"[object Opera]"==o.call(a.opera),l=!!b.attachEvent&&!l,u=r?"object":l?"script":"img",v=l?"script":u,w=Array.isArray||function(a){return"[object Array]"==o.call(a)},x=[],y={},z={timeout:function(a,b){return b.length&&(a.timeout=b[0]),a}},A,B;B=function(a){function b(a){var a=a.split("!"),b=x.length,c=a.pop(),d=a.length,c={url:c,origUrl:c,prefixes:a},e,f,g;for(f=0;f<d;f++)g=a[f].split("="),(e=z[g.shift()])&&(c=e(c,g));for(f=0;f<b;f++)c=x[f](c);return c}function g(a,e,f,g,h){var i=b(a),j=i.autoCallback;i.url.split(".").pop().split("?").shift(),i.bypass||(e&&(e=d(e)?e:e[a]||e[g]||e[a.split("/").pop().split("?")[0]]),i.instead?i.instead(a,e,f,g,h):(y[i.url]?i.noexec=!0:y[i.url]=1,f.load(i.url,i.forceCSS||!i.forceJS&&"css"==i.url.split(".").pop().split("?").shift()?"c":c,i.noexec,i.attrs,i.timeout),(d(e)||d(j))&&f.load(function(){k(),e&&e(i.origUrl,h,g),j&&j(i.origUrl,h,g),y[i.url]=2})))}function h(a,b){function c(a,c){if(a){if(e(a))c||(j=function(){var a=[].slice.call(arguments);k.apply(this,a),l()}),g(a,j,b,0,h);else if(Object(a)===a)for(n in m=function(){var b=0,c;for(c in a)a.hasOwnProperty(c)&&b++;return b}(),a)a.hasOwnProperty(n)&&(!c&&!--m&&(d(j)?j=function(){var a=[].slice.call(arguments);k.apply(this,a),l()}:j[n]=function(a){return function(){var b=[].slice.call(arguments);a&&a.apply(this,b),l()}}(k[n])),g(a[n],j,b,n,h))}else!c&&l()}var h=!!a.test,i=a.load||a.both,j=a.callback||f,k=j,l=a.complete||f,m,n;c(h?a.yep:a.nope,!!i),i&&c(i)}var i,j,l=this.yepnope.loader;if(e(a))g(a,0,l,0);else if(w(a))for(i=0;i (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];var j=d.createElement(s);var dl=l!='dataLayer'?'&l='+l:'';j.src='//www.googletagmanager.com/gtm.js?id='+i+dl;j.type='text/javascript';j.async=true;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-P4HH5NV'); Skip to main content Home About Submit ALERTS / RSS Search for this keyword Advanced Search Efficient Chest X-Ray Feature Extraction and Feature Fusion for Pneumonia Detection Using Lightweight Pretrained Deep Learning Models View ORCID Profile Yashvi Chandola , View ORCID Profile Vivek Uniyal , View ORCID Profile Yamini Bachheti doi: https://doi.org/10.1101/2025.06.28.25330459 Yashvi Chandola 1 Assistant Professor, Institute of Technology Gopeshwar , Uttarakhand Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Yashvi Chandola For correspondence: yashvi.chandola{at}gmail.com Vivek Uniyal 1 Assistant Professor, Institute of Technology Gopeshwar , Uttarakhand Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Vivek Uniyal Yamini Bachheti 2 Consulting Oncologist, Holy Care Hospital , Rishikesh, Uttarakhand Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Yamini Bachheti Abstract Full Text Info/History Metrics Data/Code Preview PDF Abstract Pneumonia is a respiratory condition characterized by inflammation of the alveolar sacs in the lungs, which disrupts normal oxygen exchange. This disease disproportionately impacts vulnerable populations, including young children (under five years of age) and elderly individuals (over 65 years), primarily due to their compromised immune systems. The mortality rate associated with pneumonia remains alarmingly high, particularly in low-resource settings where healthcare access is limited. Although effective prevention strategies exist, pneumonia continues to claim the lives of approximately one million children each year, earning its reputation as a "silent killer." Globally, an estimated 500 million cases are documented annually, underscoring its widespread public health burden. This study explores the design and evaluation of the CNN-based Computer-Aided Diagnostic (CAD) systems with an aim of carrying out competent as well as resourceful classification and categorization of chest radiographs into binary classes (Normal, Pneumonia). An augmented Kaggle dataset of 18,200 chest radiographs, split between normal and pneumonia cases, was utilized. This study conducts a series of experiments to evaluate lightweight CNN models—ShuffleNet, NASNet-Mobile, and EfficientNet-b0—using transfer learning that achieved accuracy of 90%, 88% and 89%, prompting the task for deep feature extraction from each of the networks and applying feature fusion to further pair it with SVM classifier and XGBoost classifier, achieving an accuracy of 97% and 98% resepectively. The proposed research emphasizes the crucial role of CAD systems in advancing radiological diagnostics, delivering effective solutions to aid radiologists in distinguishing between diagnoses by applying feature fusion, feature selection along with various machine learning algorithms and deep learning architectures. 1. Introduction Pneumonia is an inflammatory respiratory condition triggered by pathogenic microorganisms, including bacterial agents like Streptococcus pneumoniae, viral particles, or, in rare instances, fungal or parasitic organisms. The infection causes fluid accumulation in the alveoli, impairing gas exchange and leading to respiratory distress. While it affects all age groups, immunocompromised populations— notably children under five years old and elderly adults over 65—face the highest mortality risk. Global health data reveals alarming statistics, with approximately 15 fatalities per 100,000 individuals annually, predominantly in low-resource settings where pediatric cases account for nearly one million preventable deaths each year. With over half a billion reported cases worldwide, pneumonia has earned the grim moniker "the silent killer" due to its insidious progression and diagnostic challenges. Although no universal cure exists, preventive measures like pneumococcal conjugate vaccines (PCV) and Haemophilus influenzae type b (Hib) immunizations have significantly reduced incidence rates. Despite these advances, lower respiratory infections rank as the third-leading cause of death globally, with pneumonia representing the most severe subset. Accurate diagnosis traditionally relies on chest X-ray analysis, yet even experienced radiologists may struggle to differentiate bacterial from viral etiologies, potentially delaying appropriate treatment. In resource-constrained regions, the scarcity of specialized healthcare professionals underscores the critical need for computer-aided detection (CAD) systems. Such technologies can augment diagnostic precision, facilitate early intervention, and ultimately reduce mortality rates by supporting clinical decision-making. [ 1 – 3 ]. Deep learning is an advanced subset of artificial intelligence and machine learning that utilizes complex neural networks with multiple processing layers. These architectures are designed to replicate human cognitive functions by creating hierarchical data representations. By progressively analyzing raw input data through successive layers, deep learning models can identify and refine intricate patterns, closely mimicking human perceptual and analytical processes. Among these architectures, Convolutional Neural Networks (CNNs) are widely used for tasks like speech recognition and object detection. Typically, deep learning models are structured into input, hidden, and output layers, with each layer responsible for feature extraction and data transformation to support subsequent learning stages. In healthcare, deep learning plays a pivotal role in medical imaging by automating disease diagnosis through internal structure visualization. Imaging modalities such as X-rays, CT scans, and ultrasound generate vast datasets that can challenge manual interpretation by radiologists. Deep learning streamlines this process, enabling rapid, precise, and scalable image analysis—key advantages for timely diagnosis and improved patient management. Chest X-rays serve as a critical, non-invasive diagnostic tool for pulmonary and cardiac conditions, including pneumonia. While they involve lower radiation exposure than CT scans (which provide 3D imaging but require higher radiation doses and contrast agents), X-rays remain indispensable for detecting pneumonia-specific features like pulmonary opacities. Deep learning enhances X-ray diagnostics by efficiently processing large imaging datasets, alleviating radiologists’ workloads, and boosting diagnostic precision. This integration of AI and healthcare facilitates earlier disease detection and optimizes clinical outcomes. Pneumonia, in particular, manifests radiologically as localized or lobar pulmonary opacities, which deep learning models can reliably identify. Figure 1 (a) displays chest radiographs with a clean and transparent lung portion hence representig normal chest, whereas Figure 1.1 (b) shows chest radiographs with clouded airspace or a radio opaque lung section, clearly representing lungs infected with pneuonia. Download figure Open in new tab Fig. 1.1 Illustrative represenatation of Chest radiographs (a) A Radiograph showing Normal chest, (b) A Radiograph showing Pneumonia Infected chest 2. Literature Review The review of many Computer-Aided Diagnostic (CAD) systems established throughout the eons by researchers for 2-class categorization as well as multiple class classification of chest radiographs has been expanded in the following sections. 2.1 Computer-Aided Diagnostic (CAD) system designs for Chest Radiographs with focus on Machine Learning based algorithms Recent advancements in medical image analysis have demonstrated the effectiveness of machine learning techniques for pneumonia detection in chest X-rays. Chandra and Verma (2020) [ 26 ] proposed an innovative approach leveraging lung region-of-interest (ROI) segmentation to improve classification accuracy between normal and pneumonia cases. Their comparative analysis revealed that segmentation significantly enhanced performance across classifiers, with logistic regression achieving the highest accuracy (95.63%) when using ROI-based processing, compared to 91.50% without segmentation. Other classifiers like multi-layer perceptron (95.38%) and random forest (94.41%) also showed notable improvements with segmentation. Further supporting these findings, Al Mamlook et al. (2020) [ 27 ] reported even higher accuracy rates, with random forest reaching 97.61% and CNN achieving 98.46%, underscoring the potential of deep learning in this domain. Earlier studies by Oliveira et al. (2008) [ 28 ] and Sousa et al. (2013) [ 29 ] explored traditional machine learning methods, with Oliveira employing Haar wavelet transforms for feature extraction in a k-NN classifier, while Sousa compared SVM (77%), k-NN (70%), and Naïve Bayes (68%), later improving results using dimensionality reduction techniques like kernel-PCA. Depeursinge et al. (2010) [ 30 ] and Yao et al. (2011) [ 31 ] further validated SVM’s utility, reporting accuracies of 88.30% and 80.00%, respectively. Naydenova et al. (2015) [ 32 ] advanced this work by integrating feature selection with ensemble methods, achieving 97.80% accuracy using a hybrid random forest-SVM approach. Recent studies have introduced more sophisticated techniques. For instance, Ahmed et al. (2023) [ 121 ] combined EfficientNetV2 with XGBoost, achieving 95.2% accuracy while emphasizing model interpretability through SHAP values. Similarly, Gupta et al. (2024) [ 124 ] proposed a lightweight MobileNetV3-XGBoost pipeline for tuberculosis detection, attaining 92.8% accuracy with reduced computational costs—a critical consideration for clinical deployment. Notably, Tseng and Tang (2023) [ 123 ] optimized XGBoost with feature selection for brain tumor detection, demonstrating its versatility beyond pulmonary diseases. While prior research predominantly relied on SVM and basic classifiers. The current study addresses this gap by integrating feature fusion with SVM and XGBoost, combining the strengths of traditional and ensemble methods. This approach not only builds on existing segmentation and feature extraction strategies but also leverages XGBoost’s superior handling of imbalanced medical data, as evidenced by recent hybrid models [ 121 – 124 ]. 2.2 Computer-Aided Diagnostic (CAD) system designs for Chest Radiographs with focus on Deep Learning based CNN networks Recent advances in deep learning have revolutionized computer-aided diagnosis (CAD) systems for pulmonary conditions. Zech et al. (2018) [ 2 ] pioneered a robust classification framework using DenseNet121 with softmax classifier, validating their approach on a comprehensive multi-institutional dataset from NIH, Mount Sinai, and Indiana University networks. This multicenter study demonstrated exceptional generalization capabilities, setting a benchmark for subsequent research. Parallel work by Mubarok et al. (2019) [ 7 ] developed an assistive diagnostic system achieving 85.60% accuracy with ResNet50, though Mask R-CNN showed slightly lower performance (78.06%). Contemporary studies have pushed accuracy boundaries further. Rahman et al. (2020) [ 8 ] conducted extensive evaluations across four architectures, with DenseNet201 (98.00%) outperforming ResNet18 (96.40%) and SqueezeNet (96.10%). These findings were corroborated by Elasnaoui and Chawki (2020a) [ 15 ], whose comparative analysis of nine architectures revealed MobileNetV2 (96.27%) and ResNet50 (96.61%) as top performers. The field has particularly benefited from benchmark datasets like Kermany et al. (2018) [ 11 ]’s Kaggle collection (5,856 images), which enabled their InceptionV3 model to achieve 92.08% accuracy. Recent innovations have introduced novel architectural modifications. Liang and Zeng (2020) [ 20 ] addressed critical overfitting challenges through residual structures and dilated convolutions, while Togacar et al. (2019) [ 4 ] demonstrated exceptional performance (99.41% accuracy) combining mRMR feature selection with LDA classification. The trend toward lightweight models is evident in Chandola et al. (2021) [ 103 ]’s work, where MobileNetV2 achieved 94% accuracy, improving to 95% with decision fusion. While existing studies show promise, three key limitations emerge, most systems focus on binary classification that have limited exploration of feature fusion techniques. Khan et al. (2022) [ 126 ] developed a triple-attention CNN with 98.6% accuracy for pediatric pneumonia. Our current work introduces novel feature fusion mechanisms combining radiomic features with deep learning embeddings. The evolution from basic CNNs to sophisticated hybrid systems demonstrates remarkable progress, yet opportunities remain for improving generalizability across diverse populations and clinical settings 3. Proposed CAD System Design for Chest Radiographs The proposed Computer-Aided Diagnostic (CAD) system for pneumonia detection in chest X-rays employs a multi-stage pipeline beginning with dataset acquisition, resizing, and augmentation to ensure robust model training. It explores two complementary approaches: Lightweight end-to-end CNNs (ShuffleNet, NASNet-Mobile, EfficientNet-b0) CAD system design Hybrid CNN-ML CAD system designs where pre-trained networks (ShuffleNet, NASNet-Mobile, EfficientNet-b0) extract features for classifiers like SVM and XGBoost The proposed work involves a sequence of experiments that also involve uniquely integrating feature across multiple models (ShuffleNet, NASNet-Mobile, EfficientNet-b0) to enhance reliability, while SVM and XGBoost visualizations provide interpretability for clinical trust. Designed for both high accuracy and scalability, it addresses key gaps in existing systems—balancing computational efficiency (critical for low-resource settings) with diagnostic precision, ultimately aiming to reduce radiologist workload and improve early pneumonia detection through automated, explainable AI. The Fig. 3.1 shows the Illustrative representation of the flow of the proposed work. Download figure Open in new tab Fig. 3.1 Illustrative Representation of the flow of Proposed work 3.1 Dataset Description The experimental dataset for this study was obtained from the publicly available Kaggle repository originally curated by Kermany et al. (2018) [ 38 , 42 ]. For balanced experimental evaluation, we utilized a subset of 200 annotated chest X-ray images, with precisely 100 samples representing normal pulmonary cases and 100 demonstrating radiographic evidence of pneumonia. This carefully selected 1:1 class ratio ensures equitable representation of both diagnostic categories in our analytical framework. The dataset’s public accessibility facilitates reproducibility while its clinically validated annotations provide reliable ground truth for model training and validation. For the proposed work the images are downsized while maintaining their aspect ratio in order to maintain the lung form. Both normal and pneumonia chest radiographs are subjected to the identical enhancement strategies in this investigation. The training dataset is expanded to 18,100 radiographs from its original 100 images. First, each image undergoes translation and rotation following a horizontal and vertical flip of these translated and rotated radiographs. A total of 181 images are created by augmenting each image. The research employs a systematic data partitioning strategy, segmenting the complete dataset into three mutually exclusive groups: training, validation, and test sets. The division process begins with random allocation of 50% of the original images to the test set. The remaining half is then subjected to augmentation procedures designed to maintain balanced representation between normal and pathological cases. From this augmented pool, a stratified 10% sample is extracted to serve as the validation set. Notably, both original images and their augmented counterparts are consistently assigned to identical subsets throughout this partitioning process, preserving data integrity and preventing information leakage between sets. 3.2 Lightweight Deep Learning based CNN Networks In modern medicine, Computer-Aided Diagnostic (CAD) systems utilizing Convolutional Neural Networks (CNNs) have become essential, particularly in medical imaging analysis [ 28 – 40 ]. CNNs employ a structured, multi-layered design to identify complex patterns within visual data. These layers—comprising Convolution, Activation, Pooling, Fully Connected, and Softmax—use specialized filters to detect spatial or temporal relationships in images [ 52 – 54 ]. CNNs acquire the ability to recognize local or global features that aid in picture classification during training. To achieve optimal performance in specific tasks, CNNs can be trained effectively using various approaches. One such method is transfer learning, where pre-trained weights from a base network (initially trained on a source dataset) are applied to a target network, enabling quicker adaptation to a new dataset. Another technique is fine-tuning, which involves modifying the parameters of a pre-trained model to suit a different dataset. However, this approach is less frequently utilized in medical image analysis because of the limited availability of extensive datasets. Additionally, training the CNN from scratch is another viable option for obtaining tailored results. It can take days or weeks to train a CNN from scratch, especially when dealing with huge datasets. This can be somewhat alleviated using pre-trained networks, trained on huge benchmark datasets such as ImageNet [ 25 – 27 ]. In proposed work, transfer learning is applied, utilizing a pre-existing lightweight CNN model for the 2-class classification of chest radiographs. 3.2.1 ShuffleNet ShuffleNet, introduced by Xiangyu Zhang et al., [ 116 ] in the paper titled "ShuffleNet: An Extremely Efficient Convolutional Neural Network for Mobile Devices" is precisely understood as a lightweight CNN model specifically designed for mobile devices that lack or have limited resources primarily for computation purposes. The chief goal of ShuffleNet is to achieve high efficiency in regards of the two critical aspects which are namely computational cost (FLOPs) and memory usage, which are essential for mobile platforms. To accomplish this, ShuffleNet uses pointwise group convolution, which progresses by splitting the standard convolution into two convolution operations: depthwise convolution, where each channel that deals with the input is convolved separately, and pointwise convolution, which employs the 1x1 filter that combines the results. The network is designed to balance performance with efficiency, making it apt for real-time applications especially on mobile devices. By improving group convolutions and channel shuffling, ShuffleNet significantly reduces both memory usage and computational requirements, making it a lightweight and fast framework. 3.2.2 NASNet-Mobile NASNet-Mobile, as given by the researchers Barret Zoph et al. [ 117 ], in the paper titled "Learning Transferable Architectures for Scalable Image Recognition" is a CNN architecture designed through NAS with a focus on scalability, efficiency, and transferability across various tasks. NAS is a search process carried out automatically over a large search space to find the most effective architecture by using techniques of reinforcement learning to evolve the neural networks toward performing well for specific tasks. Thus, NASNet-Mobile was designed as a highly efficient architecture for constrained computational resources and memory on a mobile device. The core notion behind NASNet-Mobile is the use of convolutional layers along with operations that result in maximization of performance while bringing down the computation cost. Such architecture would include a stack of cells since cells are defined as the primary building blocks to be optimized for the best compromise between accuracy and efficiency for a task like image classification. NASNet-Mobile is a condensed version of NASNet architecture developed specifically for a mobile platform. NASNet-Mobile employs reduced kernel sizes with refined operational methods to achieve quite impressive performance for mobile devices endowed with limited computing power. 3.2.3 EfficientNet-b0 EfficientNet-B0, as it is described by Mingxing Tan and Quoc V. Le within the paper titled "EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks" [ 118 ], has been developed among the EfficientNet family of models to balance good performance with computing efficiency. EfficiencyNet-B0 is the smallest architecture that provides a baseline scale for other developed EfficientNet variants. The key idea of EfficientNet is the process that scales the dimensions of the network in terms of depth of the network, width of the network, and resolution of a network all at an even rate with one another, thereby improving both at the same time. Instead of just simply scaling the complexity of the network in terms of enhancing the depth of the network (by adding more layers) or increasing the width (by adding more neurons per layer), as usual in traditional scaling methods, EfficientNet practices the application of a compound scaling method where all three dimensions, namely, depth of the network, width of the network, and resolution of the network, are scaled in such a manner that the network grows in a harmonious manner with the prime aim to maintain optimal balance between the three essential dimensions rather than scaling one in a way to degrade the rest. 3.3 Machine Learning Algorithm 3.3.1 Support Vector Machine (SVM) Support Vector Machines (SVM) represent a powerful supervised learning methodology widely applied to both classification and regression tasks in machine learning. The algorithm’s fundamental principle revolves around constructing an optimal decision boundary that maximizes the separation margin between distinct classes, thereby enhancing generalization performance on unseen data. This maximum-margin classifier operates by identifying support vectors - the most critical data points that define the boundary between classes. For complex, non-linear datasets where linear separation is impossible in the original feature space, SVM employs sophisticated kernel functions to implicitly map the input data into higher-dimensional spaces where linear separation becomes achievable. Commonly used kernel functions include the Radial Basis Function (RBF), polynomial, and sigmoid kernels, each offering unique advantages for different data characteristics [ 84 , 103 , 109 ]. SVMs demonstrate particular strength in high-dimensional spaces and cases where the number of dimensions exceeds the number of samples, making them well-suited for medical imaging applications. Additionally, their inherent resistance to overfitting, especially when combined with appropriate regularization, contributes to reliable performance across diverse datasets. The algorithm’s mathematical foundation in statistical learning theory provides strong theoretical guarantees about its generalization capabilities, further reinforcing its popularity in both academic research and practical implementations. 3.3.2 XGBoost XGBoost (eXtreme Gradient Boosting) represents a powerful advancement in machine learning, employing an ensemble approach that builds upon gradient-boosted decision trees [ 119 – 121 ]. This algorithm operates through an iterative process where successive decision trees are constructed to rectify the residual errors of preceding models, progressively refining predictions through gradient descent optimization. A distinctive characteristic of XGBoost lies in its sophisticated regularization framework, incorporating both L1 (Lasso) and L2 (Ridge) penalties to control model complexity and prevent overfitting - a critical feature when working with medical imaging data where generalization is paramount. The implementation efficiently leverages parallel computing capabilities to accelerate model training while maintaining computational resource efficiency. Notably, XGBoost incorporates native functionality to handle missing values intelligently, automatically learning appropriate imputation strategies during the training process. These combined attributes - regularization, parallelization, and robust data handling - render XGBoost particularly effective for processing large-scale datasets while maintaining model interpretability. The algorithm’s inherent scalability allows it to accommodate diverse data types and sizes without compromising performance, making it adaptable to various problem domains. Furthermore, XGBoost includes built-in cross-validation support and early stopping mechanisms, enabling automated optimization of training iterations to maximize predictive accuracy while preventing unnecessary computation. This combination of technical features has established XGBoost as a preferred choice for competitive machine learning applications and real-world implementations where both accuracy and efficiency are prioritized. The algorithm’s flexibility extends to supporting various objective functions and evaluation metrics, allowing customization for specific use cases such as medical diagnosis tasks [ 120 – 126 ]. 3.4 Feature Extraction Feature extraction essentially deals with the vital task of extracting and mining substantially noteworthy information from the input image dataset. One major issue is that many variables are needed to effectively describe extracted features in huge datasets, this need of large amount of variable tends to be expensive in terms of both cost and computation time. Learning-based techniques or manual methods can be used for feature extraction. The proposed work focuses on extraction of deep features from the average pool layer of the three (ShuffleNet, NASNet-Mobile, EfficientNet-b0) pre-trained lightweight networks. A total of 544 features from ShuffleNet, 1056 features from NASNet-Mobile and 1280 features from EfficientNet-b0 are extracted from the average pooling layer of each lighweight CNN model. 3.5 Feature Fusion Feature fusion is the process of combining discriminative features extracted from multiple sources such as different neural networks, layers, or modalities to create a more robust, generalized, and informative representation for machine learning tasks such as classification, detection, or segmentation. The common techniques of feature fusion include: Concatenation : Stacking features end-to-end to preserve all original information. Element-wise Sum/Average : Merging features by adding or averaging values at each dimension. Attention-based Fusion : Weighting features dynamically based on their importance for adaptive integration. The choice of fusion method depends on the task—concatenation for complementary features, sum/average for structurally similar features, and attention mechanisms for prioritizing critical features—with each method offering unique advantages in balancing information retention and model complexity. In proposed work this technique leverages the strengths of different feature extractors, such as the Global Average Pooling (GAP) layers from multiple CNNs like ShuffleNet, NASNet-Mobile, and EfficientNet-b0, to enhance model performance by integrating complementary information. The proposed work deals with deep feature extraction from each model followed by the concatenation based feature fusion to form a ultimate deep feature set of 2880 features. 3.6 Feature Selection The act of distilling an initial big quantity of unprocessed and unanalysed data into a highy controlled, practicable, useful, and smaller subset of features is poularly called as the task of feature selection. These chosen features are frequently many, and managing sizable feature sets necessitates a variety of factors. This is addressed by feature selection, which lowers the number of variables required by limiting the feature set. The keywords Feature selection and feature dimensionality reduction are the two popularly used techiques to describe the these essential tasks of selecting a useful subsection of features from the original set of features or combining features to create new ones that are more useful anf filled with information respectively. Here, the only goal is to create a renewed feature set that accurately reflect the original dataset or is close to the original set as much as possible. Feature selection approaches include two categories of strategies namely, filter-based and wrapper-based strategies. Filter-based techniques are more commonly utilized, including chi-squared tests, box plots, and correlation-based feature selection. Although wrapper-based techniques need a significant amount of computation, model-based genetic algorithms, such as GA-SVM or GA-kNN, are utilized. Among the various feature selection techniques the proposed work focuses on correlation-based feature selection. The deep feature set of a total of 2880 features formed after fetaure fusion further undergoes corelation based feature selection resulting in a reduced feature set of 266 features. The popularly used feature selection techniques are shown in Fig. 3.2 . Download figure Open in new tab Fig. 3.2 Feature selection methods 4. Experiments Figure 4.1 Illustrates the experimental procedure pf the proposed work for classifying chest radiographs into normal and pneumonia classes. Download figure Open in new tab Fig. 4.1 Experimental procedure used in proposed work The experiments conducted in the proposed work of lightweight CNN-based deep feature extraction, feature fusion CAD system design for chest radiographs are briefly described in Table 4.1 . View this table: View inline View popup Download powerpoint Table 4.1. Overview of the Conducted Experiments in the Proposed Study Experiment 1: Design of CAD system using pre-trained Light Weight ShuffleNet CNN model This study focuses on developing a computer-aided diagnosis (CAD) system based on a pre-trained lightweight ShuffleNet CNN architecture. To improve classification accuracy, the model was trained on an augmented dataset of chest X-ray images. The performance of this CAD system, designed for binary classification of chest radiographs, is presented in Table 4.2 . View this table: View inline View popup Download powerpoint Table 4.2 Performance evaluation of lightweight pre-trained ShuffleNet CNN model Experiment 2: Design of CAD system using pre-trained Light Weight NASNet-Mobile CNN model This study focuses on developing a computer-aided diagnosis (CAD) system based on the pre-trained lightweight NASNet-Mobile convolutional neural network (CNN). To improve classification accuracy, the model was trained on an augmented dataset of chest X-ray images. The performance of this CAD system, designed for binary classification of chest radiographs, is presented in Table 4.3 . View this table: View inline View popup Download powerpoint Table 4.3 Performance evaluation of lightweight pre-trained NASNet-Mobile CNN model Experiment 3: Design of CAD system using pre-trained Light Weight EfficientNet-b0 CNN model This study focuses on developing a computer-aided diagnosis (CAD) system based on the pre-trained lightweight EfficientNet-B0 convolutional neural network (CNN). The model was trained on an augmented dataset of chest X-ray images to improve classification accuracy. Table 4.4 presents the performance metrics of the CAD system in binary classification of chest radiographs using the EfficientNet-B0 architecture. View this table: View inline View popup Download powerpoint Table 4.4 Performance evaluation of lightweight pre-trained EfficientNet-b0 CNN model Experiment 4: Design of CAD system using Deep feature extraction by Light Weight CNN models along with Feature fusion and SVM Classifier. This study focuses on developing a computer-aided diagnosis (CAD) system by applying deep feature extraction on the models trained from experiment 1 to 3. A total of 544 features from ShuffleNet, 1056 features from NASNet-Mobile and 1280 features from EfficientNet-b0 are extracted from the average pooling layer of each lighweight CNN model resulting in a deep feature set of a total of 2880 features i.e DFS = 2880, which further undergoes corelation based feature selection resulting in a reduced feature set of 266 features, i.e RFS = 266. Table 4.5 presents the performance evaluation results of the CAD system implemented using Deep feature extraction by Light Weight CNN models along with Feature fusion and SVM Classifier. View this table: View inline View popup Download powerpoint Table 4.5 Performance Evaluation of CAD System Designed Using Deep feature extraction by Light Weight CNN models along with Feature fusion and SVM Classifier Experiment 5: Design of CAD system using Deep feature extraction by Light Weight CNN models along with Feature fusion and XGBoost Classifier This study focuses on developing a computer-aided diagnosis (CAD) system by applying deep feature extraction on the models trained from experiment 1 to 3. From each lightweight CNN models average pooling layer the features are extracted resulting in a DFS of 2880 features, which further undergoes corelation based feature selection resulting in a RFS of 266 features. The results of performance evaluation of the CAD system designed using Deep feature extraction by Light Weight CNN models along with Feature fusion and XGBoost Classifier is shown in Table 4.6 . View this table: View inline View popup Download powerpoint Table 4.6 Performance Evaluation of CAD System Designed Using Deep feature extraction by Light Weight CNN models along with Feature fusion and XGBoost Classifier 5. Result & Discussions The comparative evaluation demonstrates that while individual lightweight CNN model ShuffleNet with 90% accuracy achieve respectable performance, the highest diagnostic efficacy is attained through feature fusion of multiple CNNs ShuffleNet + NASNet-Mobile + EfficientNet-b0 combined with XGBoost, achieving 98% balanced accuracy with 98% for both Normal and Pneumonia classes. This hybrid approach outperforms single-model CNNs and SVM-based fusion with 97% accuracy, highlighting that integrating complementary deep features with ensemble classifiers optimally balances sensitivity which deals with Pneumonia class detection and specificity which deals with Normal class identification, making it the most clinically reliable CAD system for chest X-ray classification. The results emphasize the superiority of feature diversity and XGBoost’s discriminative power in medical image analysis. These results align with prior studies emphasizing the role of feature optimization in improving CAD system performance [ 101 – 115 ]. However, the current work advances the field by integrating lightweight CNNs with statistical feature selection and feature extraction, offering a balance between accuracy and efficiency as shown in Table 5.1 . View this table: View inline View popup Download powerpoint Table 5.1 Comparative Evaluation of CAD System Implementations for Binary Classification of Chest X-ray Images The Fig. 5.1 illustrateds the ROC curve with its corresponding AUC values for the CAD system designed in the proposed work. Download figure Open in new tab Fig. 5 The ROC curve with its corresponding AUC values for the CAD system designed in the proposed work 6. Conclusion This study presented an efficient Computer-Aided Diagnosis system for pneumonia detection using chest X-ray images, leveraging the lightweight CNN models as deep feature extractors then further applying feature fusion and corelation based feature selection along with SVM classifier as well as XGBoost classifier. The experimental results demonstrated that the proposed model of CAD system using Deep feature extraction by Light Weight CNN models along with Feature fusion and XGBoost Classifier achieved 98% classification accuracy, highlighting its potential for accurate and rapid medical image analysis. Funding This research received no external funding. The study did not involve financial support from government agencies, private corporations, or other third-party organizations. The absence of external funding ensures that the research outcomes remain free from potential sponsor-related biases, which is particularly important in medical AI studies where commercial interests could influence methodological choices or interpretations. Conflicts of Interest The authors declare no conflicts of interest. This includes no financial, professional, or personal relationships that could be perceived as influencing the research. Data Availability All data produced are available online at https://www.kaggle.com/datasets/andrewmvd/pediatric-pneumonia-chest-xray https://www.kaggle.com/datasets/andrewmvd/pediatric-pneumonia-chest-xray References [1]. ↵ World Health Organization , 2014 . Revised WHO classification and treatment of pneumonia in children at health facilities: evidence summaries . [2]. ↵ Sánchez , Y.G.R. , Pérez , A.L.S. and Zacarias , I.M ., 2018 . Mortalidad por causas en el estado de México, 2000 y 2015/Mortality by causes in the state of Mexico, 2000 and 2015 . Novedades en Población , 14 ( 28 ), pp. 64 – 78 . OpenUrl [3]. ↵ World Health Organization , 2001 . Standardization of interpretation of chest radiographs for the diagnosis of pneumonia in children (No. WHO/V&B/01.35) . World Health Organization . [4]. ↵ Abdallah Y , Mohamed S . Automatic recognition of leukemia cells using texture analysis algorithm . International Journal of Advanced Research (IJAR ). 2016 ; 4 ( 1 ): 1242 – 1248 OpenUrl [5]. J. Islam and Y. Zhang , ‘Brain MRI analysis for Alzheimer’s disease diagnosis using an ensemble system of deep convolutional neural networks,’ Brain Informatics , Vol. 5 , No. 2 , 2018a . [6]. S. Sarraf , G. Tofighi , ’Classification of Alzheimer’s disease using fMRI data and deep learning convolutional neural networks’ , Computer Science Conference Proceedings , pp. 109 – 119 , 2016 . [7]. ↵ Cruz-Roa , A. , Gilmore , H. , Basavanhally , A. , Feldman , M. , Ganesan , S. , Shih , N.N. , Tomaszewski , J. , González , F.A. and Madabhushi , A ., 2017 . Accurate and reproducible invasive breast cancer detection in whole-slide images: A Deep Learning approach for quantifying tumor extent . Scientific reports , 7 , p. 46450 . OpenUrl PubMed [8]. ↵ H. Suk , D. Shen , ‘ Deep learning-based feature representation for AD/MCI classification ’, Inproceedings of Medical Image Computing and Computer-Assisted Intervention , pp. 583 – 590 , 2013 . [9]. Pereira , J.B. , Mijalkov , M. , Kakaei , E. , Mecocci , P. , Vellas , B. , Tsolaki , M. , Kłoszewska , I. , Soininen , H. , Spenger , C. , Lovestone , S. and Simmons , A ., 2016 . Disrupted network topology in patients with stable and progressive mild cognitive impairment and Alzheimer’s disease . Cerebral Cortex , 26 ( 8 ), pp. 3476 – 3493 . OpenUrl CrossRef PubMed [10]. A. Ortiz , J. Munilla , ‘Ensembles of deep learning architectures for the early diagnosis of the Alzheimer’s disease’ , International Journal of Neural Systems , Vol. 26 , No. 7 , pp. 1650025-1-1650025-23, 2016 . [11]. ↵ J. Rieke , F. Eitel , M. Weygandt , J.D. Haynes , K. Ritter ‘ Visualizing convolutional networks for MRI-based diagnosis of Alzheimer’s disease ’, Alzheimer’s disease Neuroimaging Initiative , 2018 . [12]. Kriti , Virmani , J. and Agarwal , R. , 2019 . Effect of despeckle filtering on classification of breast tumors using ultrasound images . Biocybernetics and Biomedical Engineering , 39 ( 2 ), pp. 536 – 560 . OpenUrl [13]. Kriti , Virmani , J. and Agarwal , R. , 2019 . Assessment of despeckle filtering algorithms for segmentation of breast tumours from ultrasound images . Biocybernetics and Biomedical Engineering , 39 ( 1 ), pp. 100 – 121 . OpenUrl [14]. Ashburner , J. and Friston , K.J ., 2000 . Voxel-based morphometry—the methods . Neuroimage , 11 ( 6 ), pp. 805 – 821 . OpenUrl CrossRef PubMed Web of Science [15]. ↵ Ricci , E. and Perfetti , R ., 2007 . Retinal blood vessel segmentation using line operators and support vector classification . IEEE transactions on medical imaging , 26 ( 10 ), pp. 1357 – 1365 . OpenUrl CrossRef PubMed [16]. Baldauf , C. , Bäuerle , A. , Ropinski , T. , Rasche , V. and Vernikouskaya , I. , 2019 . Convolutional neural network (CNN) applied to respiratory motion detection in fluoroscopic frames (Doctoral dissertation, Ulm University). [17]. Cervino , L.I. , Du , J. and Jiang , S.B ., 2011 . MRI-guided tumor tracking in lung cancer radiotherapy . Physics in Medicine & Biology , 56 ( 13 ), p. 3773 . OpenUrl PubMed [18]. Kamnitsas , K. , Ledig , C. , Newcombe , V.F. , Simpson , J.P. , Kane , A.D. , Menon , D.K. , Rueckert , D. and Glocker , B. , 2017 . Efficient multi-scale 3D CNN with fully connected CRF for accurate brain lesion segmentation . Medical image analysis , 36 , pp. 61 – 78 . OpenUrl CrossRef PubMed [19]. Li , J. , Fan , M. , Zhang , J. and Li , L ., 2017 , March. Discriminating between benign and malignant breast tumors using 3D convolutional neural network in dynamic contrast enhanced-MR images . In Medical Imaging 2017: Imaging Informatics for Healthcare, Research, and Applications (Vol. 10138 , p. 1013808 ). International Society for Optics and Photonics . OpenUrl [20]. ↵ M. Colombo , and G. Ronchi , “Focal Liver Lesions – Detection, Characterization, Ablation”, Springer , Berlin , pp. 167 – 177 , 2005 . [21]. Hsu , W.H. , Tsai , F.J. , Zhang , G. , Chang , C.K. , Hsieh , P.H. , Yang , S.N. , Sun , S.S. , Liao , K. and Huang , E.T ., 2019 . Development of a Deep Learning Model for Chest X-Ray Screening . MEDICAL PHYSICS INTERNATIONAL , 7 ( 3 ), p. 314 . OpenUrl [22]. Stadler , J.A. , Andronikou , S. and Zar , H.J ., 2017 . Lung ultrasound for the diagnosis of community-acquired pneumonia in children . Pediatric radiology , 47 ( 11 ), pp. 1412 – 1419 . OpenUrl PubMed [23]. Amundsen , T. , Torheim , G. , Waage , A. , Bjermer , L. , Steen , P.A. and Haraldseth , O ., 2000 . Perfusion magnetic resonance imaging of the lung: characterization of pneumonia and chronic obstructive pulmonary disease . A feasibility study. Journal of Magnetic Resonance Imaging: An Official Journal of the International Society for Magnetic Resonance in Medicine , 12 ( 2 ), pp. 224 – 231 . OpenUrl [24]. Patel , V.K. , Naik , S.K. , Naidich , D.P. , Travis , W.D. , Weingarten , J.A. , Lazzaro , R. , Gutterman , D.D. , Wentowski , C. , Grosu , H.B. and Raoof , S ., 2013 . A practical algorithmic approach to the diagnosis and management of solitary pulmonary nodules: part 1: radiologic characteristics and imaging modalities . Chest , 143 ( 3 ), pp. 825 – 839 . OpenUrl CrossRef PubMed [25]. ↵ Rosman , D.A. , Duszak Jr , R. , Wang , W. , Hughes , D.R. and Rosenkrantz , A.B ., 2018 . Changing utilization of noninvasive diagnostic imaging over 2 decades: an examination family–focused analysis of medicare claims using the Neiman imaging types of service categorization system . American Journal of Roentgenology , 210 ( 2 ), pp. 364 – 368 . OpenUrl PubMed [26]. ↵ Ash , S.Y. , Estépar , R.S.J. and Washko , G.R. , 2020 . Chest Imaging for Precision Medicine. In Precision in Pulmonary, Critical Care, and Sleep Medicine (pp. 107 – 115 ). Humana , Cham . [27]. ↵ Tierney , D.M. , Huelster , J.S. , Overgaard , J.D. , Plunkett , M.B. , Boland , L.L. , Hill , C.A.S. , Agboto , V.K. , Smith , C.S. , Mikel , B.F. , Weise , B.E. and Madigan , K.E ., 2020 . Comparative performance of pulmonary ultrasound, chest radiograph, and CT among patients with acute respiratory failure . Critical care medicine , 48 ( 2 ), pp. 151 – 157 . OpenUrl PubMed [28]. ↵ Gu , X. , Pan , L. , Liang , H. and Yang , R. , 2018 , March. Classification of bacterial and viral childhood pneumonia using deep learning in chest radiography . In Proceedings of the 3rd International Conference on Multimedia and Image Processing (pp. 88 – 93 ). [29]. ↵ Zech , J.R. , Badgeley , M.A. , Liu , M. , Costa , A.B. , Titano , J.J. and Oermann , E.K ., 2018 . Variable generalization performance of a deep learning model to detect pneumonia in chest radiographs: a cross-sectional study . PLoS medicine , 15 ( 11 ), p. e1002683 . OpenUrl [30]. ↵ Chowdhury , M.E. , Rahman , T. , Khandakar , A. , Mazhar , R. , Kadir , M.A. , Mahbub , Z.B. , Islam , K.R. , Khan , M.S. , Iqbal , A. , Al-Emadi , N. and Reaz , M.B.I ., 2020 . Can AI help in screening viral and COVID-19 pneumonia? . arXiv preprint arXiv:2003.13145. [31]. ↵ Toğaçar , M. , Ergen , B. and Cömert , Z ., 2019 . A deep feature learning model for pneumonia detection applying a combination of mRMR feature selection and machine learning models . IRBM . [32]. ↵ Li , Z. , Yu , J. , Li , X. , Li , Y. , Dai , W. , Shen , L. , Mou , L. and Pu , Z. , 2019 , October. PNet: An Efficient Network for Pneumonia Detection. In 2019 12th International Congress on Image and Signal Processing , BioMedical Engineering and Informatics (CISP-BMEI) (pp. 1 – 5 ). IEEE. [33]. Sharma , H. , Jain , J.S. , Bansal , P. and Gupta , S. , 2020 , January. Feature Extraction and Classification of Chest radiographs Using CNN to Detect Pneumonia . In 2020 10th International Conference on Cloud Computing, Data Science & Engineering (Confluence) (pp. 227 – 231 ). IEEE. [34]. Al Mubarok , A.F. , Dominique , J.A. and Thias , A.H. , 2019 , March. Pneumonia Detection with Deep Convolutional Architecture . In 2019 International Conference of Artificial Intelligence and Information Technology (ICAIIT) (pp. 486 – 489 ). IEEE. [35]. Rahman , T. , Chowdhury , M.E. , Khandakar , A. , Islam , K.R. , Islam , K.F. , Mahbub , Z.B. , Kadir , M.A. and Kashem , S ., 2020 . Transfer Learning with Deep Convolutional Neural Network (CNN) for Pneumonia Detection using Chest X-ray . Applied Sciences , 10 ( 9 ), p. 3233 . OpenUrl [36]. Elasnaoui , K. and Chawki , Y ., 2020 . Using X-ray images and deep learning for automated detection of coronavirus disease. Journal of Biomolecular Structure and Dynamics , (just-accepted), pp. 1 – 22 . [37]. Saraiva , A.A. , Ferreira , N.M.F. , de Sousa , L.L. , Costa , N.J.C. , Sousa , J.V.M. , Santos , D.B.S. , Valente , A. and Soares , S. , 2019 . Classification of Images of Childhood Pneumonia using Convolutional Neural Networks . In BIOIMAGING (pp. 112 – 119 ). [38]. ↵ Kermany , D.S. , Goldbaum , M. , Cai , W. , Valentim , C.C. , Liang , H. , Baxter , S.L. , McKeown , A. , Yang , G. , Wu , X. , Yan , F. and Dong , J ., 2018 . Identifying medical diagnoses and treatable diseases by image-based deep learning . Cell , 172 ( 5 ), pp. 1122 – 1131 . OpenUrl CrossRef PubMed [39]. Jakhar , K. and Hooda , N. , 2018 , December. Big Data Deep Learning Framework using Keras: A Case Study of Pneumonia Prediction . In 2018 4th International Conference on Computing Communication and Automation (ICCCA) (pp. 1 – 5 ). IEEE. [40]. ↵ Rajaraman , S. , Candemir , S. , Kim , I. , Thoma , G. and Antani , S ., 2018 . Visualization and interpretation of convolutional neural network predictions in detecting pneumonia in pediatric chest radiographs . Applied Sciences , 8 ( 10 ), p. 1715 . OpenUrl PubMed [41]. Ayan , E. and Ünver , H.M. , 2019 , April. Diagnosis of Pneumonia from Chest X-Ray Images Using Deep Learning . In 2019 Scientific Meeting on Electrical-Electronics & Biomedical Engineering and Computer Science (EBBT) (pp. 1 – 5 ). IEEE. [42]. ↵ “ https://www.kaggle.com/andrewmvd/pediatric-pneumonia-chest-xray ” [43]. Gadermayr , M. , Dombrowski , A.K. , Klinkhammer , B.M. , Boor , P. and Merhof , D. , 2017 . CNN cascades for segmenting whole slide images of the kidney . arXiv preprint arXiv:1708.00251. [44]. Islam , J. , Zhang , Y . and Alzheimer’s Disease Neuroimaging Initiative, 2018 , December. Deep Convolutional Neural Networks for Automated Diagnosis of Alzheimer’s Disease and Mild Cognitive Impairment Using 3D Brain MRI . In International Conference on Brain Informatics (pp. 359 – 369 ). Springer , Cham . [45]. Makde , V. , Bhavsar , J. , Jain , S. and Sharma , P. , 2017 , March. Deep neural network based classification of tumourous and non-tumorous medical images . In International Conference on Information and Communication Technology for Intelligent Systems (pp. 199 – 206 ). Springer , Cham . [46]. Kolachalama , V.B. , Singh , P. , Lin , C.Q. , Mun , D. , Belghasem , M.E. , Henderson , J.M. , Francis , J.M. , Salant , D.J. and Chitalia , V.C ., 2018 . Association of pathological fibrosis with renal survival using deep neural networks . Kidney international reports , 3 ( 2 ), pp. 464 – 475 . OpenUrl PubMed [47]. Choi , J.W. , Ku , Y. , Yoo , B.W. , Kim , J.A. , Lee , D.S. , Chai , Y.J. , Kong , H.J. and Kim , H.C ., 2017 . White blood cell differential count of maturation stages in bone marrow smear using dual-stage convolutional neural networks . PloS one , 12 ( 12 ), p. e0189259 . OpenUrl CrossRef PubMed [48]. Islam , J. and Zhang , Y ., 2018 . Brain MRI analysis for Alzheimer’s disease diagnosis using an ensemble system of deep convolutional neural networks . Brain informatics , 5 ( 2 ), p. 2 . OpenUrl PubMed [49]. Gessert , N. , Heyder , M. , Latus , S. , Lutz , M. and Schlaefer , A. , 2018 . Plaque classification in coronary arteries from ivoct images using convolutional neural networks and transfer learning . arXiv preprint arXiv:1804.03904. [50]. Jaiswal , A.K. , Tiwari , P. , Kumar , S. , Gupta , D. , Khanna , A. and Rodrigues , J.J ., 2019 . Identifying pneumonia in chest radiographs: A deep learning approach . Measurement , 145 , pp. 511 – 518 . OpenUrl [51]. Shorten , C. and Khoshgoftaar , T.M ., 2019 . A survey on image data augmentation for deep learning . Journal of Big Data , 6 ( 1 ), p. 60 . OpenUrl [52]. ↵ Hu , B. , Lei , C. , Wang , D. , Zhang , S. and Chen , Z. , 2019 . A Preliminary Study on Data Augmentation of Deep Learning for Image Classification . arXiv preprint arXiv:1906.11887. [53]. Mikołajczyk , A. and Grochowski , M. , 2018 , May. Data augmentation for improving deep learning in image classification problem . In 2018 international interdisciplinary PhD workshop (IIPhDW) (pp. 117 – 122 ). IEEE. [54]. ↵ Oliveira , D.A.B. and Viana , M.P. , 2018 , September. Lung nodule synthesis using cnn-based latent data representation . In International Workshop on Simulation and Synthesis in Medical Imaging (pp. 111 – 118 ). Springer , Cham . [55]. Perez , L. and Wang , J. , 2017 . The effectiveness of data augmentation in image classification using deep learning . arXiv preprint arXiv:1712.04621. [56]. Hussain , Z. , Gimenez , F. , Yi , D. and Rubin , D ., 2017 . Differential data augmentation techniques for medical imaging classification tasks . In AMIA Annual Symposium Proceedings (Vol. 2017 , p. 979 ). American Medical Informatics Association . OpenUrl PubMed [57]. Shin , H.C. , Roth , H.R. , Gao , M. , Lu , L. , Xu , Z. , Nogues , I. , Yao , J. , Mollura , D. and Summers , R.M ., 2016 . Deep convolutional neural networks for computer-aided detection: CNN architectures, dataset characteristics and transfer learning . IEEE transactions on medical imaging , 35 ( 5 ), pp. 1285 – 1298 . OpenUrl CrossRef PubMed [58]. Tajbakhsh , N. , Shin , J.Y. , Gurudu , S.R. , Hurst , R.T. , Kendall , C.B. , Gotway , M.B. and Liang , J ., 2016 . Convolutional neural networks for medical image analysis: Full training or fine tuning? . IEEE transactions on medical imaging , 35 ( 5 ), pp. 1299 – 1312 . OpenUrl CrossRef PubMed [59]. Liang , G. and Zheng , L ., 2019 . A transfer learning method with deep residual network for pediatric pneumonia diagnosis . Computer methods and programs in biomedicine , p. 104964 . [60]. Yang , Q ., 2008 , October. An Introduction to Transfer Learning . In ADMA (p. 1 ). [61]. Raghu , M. , Zhang , C. , Kleinberg , J. and Bengio , S. , 2019 . Transfusion: Understanding transfer learning for medical imaging . In Advances in neural information processing systems (pp. 3347 – 3357 ). [62]. Kutlu , H. and Avcı , E ., 2019 . A novel method for classifying liver and brain tumors using convolutional neural networks, discrete wavelet transform and long short-term memory networks . Sensors , 19 ( 9 ), p. 1992 . OpenUrl PubMed [63]. Jirak , D. , Dezortová , M. , Taimr , P. and Hájek , M ., 2002 . Texture analysis of human liver . Journal of Magnetic Resonance Imaging: An Official Journal of the International Society for Magnetic Resonance in Medicine , 15 ( 1 ), pp. 68 – 74 . OpenUrl [64]. Fetit , A.E. , Novak , J. , Peet , A.C. and Arvanitis , T.N ., 2015 . Three-dimensional textural features of conventional MRI improve diagnostic classification of childhood brain tumours . NMR in Biomedicine , 28 ( 9 ), pp. 1174 – 1184 . OpenUrl CrossRef PubMed [65]. Ballerini , L. , Fisher , R.B. , Aldridge , B. and Rees , J. , 2013 . A color and texture based hierarchical K-NN approach to the classification of non-melanoma skin lesions . In Color Medical Image Analysis (pp. 63 – 86 ). Springer , Dordrecht . [66]. Nasir , M. , Attique Khan , M. , Sharif , M. , Lali , I.U. , Saba , T. and Iqbal , T ., 2018 . An improved strategy for skin lesion detection and classification using uniform segmentation and feature selection based approach . Microscopy research and technique , 81 ( 6 ), pp. 528 – 543 . OpenUrl PubMed [67]. Shijie , J. , Ping , W. , Peiyi , J. and Siping , H. , 2017 , October. Research on data augmentation for image classification based on convolution neural networks . In 2017 Chinese automation congress (CAD) (pp. 4165 – 4170 ). IEEE. [68]. Zhu , Y. , Chen , Y. , Lu , Z. , Pan , S.J. , Xue , G.R. , Yu , Y. and Yang , Q. , 2011 , August. Heterogeneous transfer learning for image classification . In Twenty-Fifth AAAI Conference on Artificial Intelligence. [69]. Weiss , K. , Khoshgoftaar , T.M. and Wang , D ., 2016 . A survey of transfer learning . Journal of Big data , 3 ( 1 ), p. 9 . OpenUrl [70]. Krizhevsky , A. , Sutskever , I. and Hinton , G.E ., 2012 . 2012 AlexNet . Adv. Neural Inf. Process. Syst ., pp. 1 – 9 . [71]. Abd Almisreb , A. , Saleh , M.A. and Tahir , N.M. , 2019 , October. Anomalous Behaviour Detection using Transfer Learning Algorithm of Series and DAG Network . In 2019 IEEE 9th International Conference on System Engineering and Technology (ICSET) (pp. 505 – 509 ). IEEE. [72]. Bhandary , A. , Prabhu , G.A. , Rajinikanth , V. , Thanaraj , K.P. , Satapathy , S.C. , Robbins , D.E. , Shasky , C. , Zhang , Y.D. , Tavares , J.M.R. and Raja , N.S.M ., 2020 . Deep-learning framework to detect lung abnormality–A study with chest X-Ray and lung CT scan images . Pattern Recognition Letters , 129 , pp. 271 – 278 . OpenUrl [73]. Nawaz , W. , Ahmed , S. , Tahir , A. and Khan , H.A. , 2018 , June. Classification of breast cancer histology images using alexnet . In International conference image analysis and recognition (pp. 869 – 876 ). Springer , Cham . [74]. He , K. , Zhang , X. , Ren , S. and Sun , J. , 2016 . Deep residual learning for image recognition . In Proceedings of the IEEE conference on computer vision and pattern recognition (pp. 770 – 778 ). [75]. Ayyachamy , S. , Alex , V. , Khened , M. and Krishnamurthi , G ., 2019 , March. Medical image retrieval using Resnet-18 . In Medical Imaging 2019: Imaging Informatics for Healthcare, Research, and Applications (Vol. 10954 , p. 1095410 ). International Society for Optics and Photonics . OpenUrl [76]. Zhu , Y. , Fu , Z. and Fei , J. , 2017 , December. An image augmentation method using convolutional network for thyroid nodule classification by transfer learning . In 2017 3rd IEEE International Conference on Computer and Communications (ICCC) (pp. 1819 – 1823 ). IEEE. [77]. Szegedy , C. , Liu , W. , Jia , Y. , Sermanet , P. , Reed , S. , Anguelov , D. , Erhan , D. , Vanhoucke , V. and Rabinovich , A. , 2015 . Going deeper with convolutions . In Proceedings of the IEEE conference on computer vision and pattern recognition (pp. 1 – 9 ). [78]. Szegedy , C. , Ioffe , S. , Vanhoucke , V. and Alemi , A. , Inception-v4, inception-resnet and the impact of residual connections on learning . arXiv 2016. arXiv preprint arXiv:1602.07261. [79]. Khagi , B. , Lee , B. , Pyun , J.Y. and Kwon , G.R. , 2019 , January. CNN Models Performance Analysis on MRI images of OASIS dataset for distinction between Healthy and Alzheimer’s patient . In 2019 International Conference on Electronics, Information, and Communication (ICEIC) (pp. 1 – 4 ). IEEE. [80]. Nixon , M. and Aguado , A. , 2019 . Feature extraction and image processing for computer vision . Academic press . [81]. Rawat , J. , Singh , A. , Bhadauria , H.S. , Virmani , J. and Devgun , J.S ., 2018b . Leukocyte Classification using Adaptive Neuro-Fuzzy Inference System in Microscopic Blood Images . Arabian Journal for Science and Engineering , 43 ( 12 ), pp. 7041 – 7058 . OpenUrl [82]. Kher , R. , Pawar , T. , Thakar , V. and Shah , H ., 2015 . Physical activities recognition from ambulatory ECG signals using neuro-fuzzy classifiers and support vector machines . Journal of medical engineering & technology , 39 ( 2 ), pp. 138 – 152 . OpenUrl PubMed [83]. Do , Q.H. and Chen , J.F ., 2013 . A neuro-fuzzy approach in the classification of students’ academic performance . Computational intelligence and neuroscience , 2013 . 84. ↵ [84] Chang , C.C. , Lin , C.J. : LIBSVM, A library of support vector machines ( 2012 ) [85]. J. Virmani , V. Kumar , N. Kalra , and N. Khandelwal , “ SVM-based characterization of liver ultrasound images using wavelet packet texture descriptors ”, J Digit Imaging , Vol. 26 , No. 3 , pp. 530 – 543 , 2013 . OpenUrl CrossRef PubMed [86]. J. Virmani , V. Kumar , N. Kalra , and N. Khandelwal , “ PCA-SVM based CAD system for focal liver lesions from B-mode ultrasound ”, Def Sci J , Vol. 63 , No. 5 , pp. 478 – 486 , 2013 . OpenUrl [87]. J. Virmani , V. Kumar , N. Kalra , and N. Khandelwal , “ Characterization of primary and secondary malignant liver lesions from B-mode ultrasound ”, J Digit Imaging , Vol. 26 , No. 6 , pp. 1058 – 1070 , 2013 . OpenUrl PubMed [88]. J. Virmani , V. Kumar , N. Kalra , and N. Khandelwal , “ Prediction of liver cirrhosis based on multiresolution texture descriptors from B-mode ultrasound ”, Int J of Convergence Comput , Vol. 1 , No. 1 , pp. 19 – 37 , 2013 . OpenUrl [89]. Hassanein , A.E. , Kim , T.H. : Breast cancer MRI diagnosis approach using support vector 90. [90] machine and pulse coupled neural networks . J. Appl. Logic 10 ( 4 ), 274 – 284 ( 2012 ) OpenUrl [91]. J. Virmani , V. Kumar , N. Kalra , and N. Khandelwal , “ Prediction of cirrhosis from liver ultrasound B-mode images based on Laws’ mask analysis ”, In: Proceedings of the IEEE Int Conf on Image Inf Processing, ICIIP-2011. Himachal Pradesh, India, pp. 1 – 5 , 2011 . [92]. J. Virmani , V. Kumar , N. Kalra , and N. Khandelwal , “ Neural network ensemble based CAD system for focal liver lesions using B-mode ultrasound ”, J Digit Imaging , Vol. 27 , No. 4), pp. 520 – 537 , 2014 . OpenUrl PubMed [93]. J. Virmani , V. Kumar , N. Kalra , and N. Khandelwal , “ A comparative study of Computer-Aided Diagnostic systems for focal hepatic lesions from B-mode ultrasound ”, Journal of Medical Engineering and Technology , Vol. 37 , No. 4 , pp. 229 – 306 , 2013 . OpenUrl [94]. J. Virmani , V. Kumar , N. Kalra , and N. Khandelwal , “ A rapid approach for prediction of liver cirrhosis based on first order statistics ”, In: Proceedings of IEEE International Conference on Multimedia, Signal Processing and Communication Technologies, IMPACT-2011, pp. 212 – 215 , 2011 . [95]. Zhang , T. , Recent Advances in Convolution Neural Networks . [96]. Vaseli , H. , Liao , Z. , Abdi , A.H. , Girgis , H. , Behnami , D. , Luong , C. , Dezaki , F.T. , Dhungel , N. , Rohling , R. , Gin , K. and Abolmaesumi , P ., 2019 , March. Designing lightweight deep learning models for echocardiography view classification . In Medical Imaging 2019: Image-Guided Procedures, Robotic Interventions, and Modeling (Vol. 10951 , p. 109510F ). International Society for Optics and Photonics . OpenUrl [97]. Iandola , F.N. , Han , S. , Moskewicz , M.W. , Ashraf , K. , Dally , W.J. and Keutzer , K. , 2016 . SqueezeNet: AlexNet-level accuracy with 50x fewer parameters and<0.5 MB model size . arXiv preprint arXiv:1602.07360. [98]. Pal , A. , Jaiswal , S. , Ghosh , S. , Das , N. and Nasipuri , M. , 2018 , December. Segfast: A faster squeezenet based semantic image segmentation technique using depth-wise separable convolutions . In Proceedings of the 11th Indian Conference on Computer Vision, Graphics and Image Processing (pp. 1 – 7 ). [99]. Ucar , F. and Korkmaz , D ., 2020 . COVIDiagnosis-Net: Deep Bayes-SqueezeNet based Diagnostic of the Coronavirus Disease 2019 (COVID-19) from X-Ray Images . Medical Hypotheses , p. 109761 . [100]. Nakamichi , K. , Lu , H. , Kim , H. , Yoneda , K. and Tanaka , F. , 2019 , October. Classification of Circulating Tumor Cells in Fluorescence Microscopy Images Based on SqueezeNet . In 2019 19th International Conference on Control, Automation and Systems (ICCAS) (pp. 1042 – 1045 ). IEEE. [101]. ↵ Qian , X. , Patton , E.W. , Swaney , J. , Xing , Q. and Zeng , T. , 2018 , October. Machine learning on cataracts classification using SqueezeNet . In 2018 4th International Conference on Universal Village (UV) (pp. 1 – 3 ). IEEE. [102]. Kriti , Virmani , J. , Dey , N. and Kumar , V. , 2016 . PCA-PNN and PCA-SVM based CAD systems for breast density classification . In Applications of intelligent optimization in biology and medicine (pp. 159 – 180 ). Springer , Cham . [103]. ↵ Chandola , Y. , Virmani , J. , Bhadauria , H.S. and Kumar , P ., 2021 . Deep Learning for Chest Radiographs: Computer-Aided Diagnostic . Elsevier . [104]. Chandola , Y. , Virmani , J. , Bhadauria , H. and Kumar , P ., 2021 . Chapter 1 - Introduction, Deep Learning for Chest Radiographs , pp. 1 – 33 ., In Primers in Biomedical Imaging Devices and Systems, ISBN 9780323901840 , doi: 10.1016/B978-0-323-90184-0.00003-5 . OpenUrl CrossRef [105]. Chandola , Y. , Virmani , J. , Bhadauria , H. and Kumar , P ., 2021 . Chapter 2 - Review of related work, Deep Learning for Chest Radiographs , pp. 35 – 57 ., In Primers in Biomedical Imaging Devices and Systems, ISBN 9780323901840 , doi: 10.1016/B978-0-323-90184-0.00010-2 . OpenUrl CrossRef [106]. Chandola , Y. , Virmani , J. , Bhadauria , H. and Kumar , P ., 2021 . Chapter 3 - Methodology adopted for designing of Computer-Aided Diagnostic systems for chest radiographs . Deep Learning for Chest Radiographs , pp. 59 – 115 , ISBN 9780323901840 , doi: 10.1016/B978-0-323-90184-0.00008-4 . OpenUrl CrossRef [107]. Chandola , Y. , Virmani , J. , Bhadauria , H. and Kumar , P ., 2021 . Chapter 4 - End-to-end pre-trained CNN-based Computer-Aided Diagnostic system design for chest radiographs . Deep Learning for Chest Radiographs , pp. 117 – 140 , ISBN 9780323901840 , doi: 10.1016/B978-0-323-90184-0.00011-4 . OpenUrl CrossRef [108]. Chandola , Y. , Virmani , J. , Bhadauria , H. and Kumar , P ., 2021 . Chapter 5 - Hybrid Computer-Aided Diagnostic system design using end-to-end CNN-based deep feature extraction and ANFC-LH classifier for chest radiographs . Deep Learning for Chest Radiographs , pp. 141 – 156 , ISBN 9780323901840 , doi: 10.1016/B978-0-323-90184-0.00007-2 . OpenUrl CrossRef [109]. ↵ Chandola , Y. , Virmani , J. , Bhadauria , H. and Kumar , P ., 2021 . Chapter 6 - Hybrid Computer-Aided Diagnostic system design using end-to-end Pre-trained CNN-based deep feature extraction and PCA-SVM classifier for chest radiographs , Deep Learning for Chest Radiographs , pp. 157 – 166 , ISBN 9780323901840 , doi: 10.1016/B978-0-323-90184-0.00005-9 . OpenUrl CrossRef [110]. Chandola , Y. , Virmani , J. , Bhadauria , H. and Kumar , P ., 2021 . Chapter 7 - Lightweight end-to- end Pre-trained CNN-based Computer-Aided Diagnostic system design for chest radiographs , Deep Learning for Chest Radiographs , pp. 167 – 183 , ISBN 9780323901840 , doi: 10.1016/B978-0-323-90184-0.00001-1 . OpenUrl CrossRef [111]. Chandola , Y. , Virmani , J. , Bhadauria , H. and Kumar , P ., 2021 . Chapter 8 - Hybrid Computer-Aided Diagnostic system design using lightweight end-to-end Pre-trained CNN-based deep feature extraction and ANFC-LH classifier for chest radiographs , Deep Learning for Chest Radiographs , pp. 185 – 196 , ISBN 9780323901840 , doi: 10.1016/B978-0-323-90184-0.00009-6 . OpenUrl CrossRef [112]. Chandola , Y. , Virmani , J. , Bhadauria , H. and Kumar , P ., 2021 . Chapter 9 - Hybrid Computer-Aided Diagnostic system design using lightweight end-to-end Pre-trained CNN-based deep feature extraction and PCA-SVM classifier for chest radiographs , Deep Learning for Chest Radiographs , pp. 197 – 204 , ISBN 9780323901840 , doi: 10.1016/B978-0-323-90184-0.00006-0 . OpenUrl CrossRef [113]. Chandola , Y. , Virmani , J. , Bhadauria , H. and Kumar , P ., 2021 . Chapter 10 - Comparative analysis of Computer-Aided Diagnostic systems designed for chest radiographs: Conclusion and future scope , Deep Learning for Chest Radiographs , pp. 205 – 209 , ISBN 9780323901840 , doi: 10.1016/B978-0-323-90184-0.00004-7 . OpenUrl CrossRef [114]. Joshi , S. , Shah , R. , Chandola , Y. , Uniyal , V ., 2024 , Classification of Brain MRI Images using End-to-end Trained AlexNet & End-to-end Pre-trained MobileNet , International Journal of Research Publication and Reviews , Vol. 5 , Issue 7 , pp 205 – 218 , New Global Publications OpenUrl [115]. ↵ Chandola , Y. , Uniyal , V. , Bachheti , Y. , Lakhera , N. , Rawat , R ., 2024 , Data Augmentation Techniques Applied to Medical Images , International Journal of Research Publication and Reviews , Vol. 5 , Issue 7 , pp 483 – 501 , New Global Publications OpenUrl [116]. ↵ Zhang , X. , Zhou , X. , Lin , M. and Sun , J. , 2018 . Shufflenet: An extremely efficient convolutional neural network for mobile devices . In Proceedings of the IEEE conference on computer vision and pattern recognition (pp. 6848 – 6856 ). [117]. ↵ Zoph , B. , Vasudevan , V. , Shlens , J. and Le , Q.V ., 2017 . Learning transferable architectures for scalable image recognition , CoRR abs/1707.07012. arXiv preprint arXiv:1707.07012, 61. [118]. ↵ Tan , M. and Le , Q. , 2019 , May. Efficientnet: Rethinking model scaling for convolutional neural networks . In International conference on machine learning (pp. 6105 – 6114 ). PMLR. [119]. ↵ Chen , T. and Guestrin , C. , 2016 , August. Xgboost: A scalable tree boosting system . In Proceedings of the 22nd acm sigkdd international conference on knowledge discovery and data mining (pp. 785 – 794 ). [120]. ↵ Nielsen , D. , 2016 . Tree boosting with xgboost-why does xgboost win" every" machine learning competition? (Master’s thesis, NTNU). [121]. ↵ Ahmed , M.S. , Rahman , A. , AlGhamdi , F. , AlDakheel , S. , Hakami , H. , AlJumah , A. , AlIbrahim , Z. , Youldash , M. , Alam Khan , M.A. and Basheer Ahmed , M.I ., 2023 . Joint diagnosis of pneumonia, COVID-19, and tuberculosis from chest X-ray images: A deep learning approach . Diagnostics , 13 ( 15 ), p. 2562 . OpenUrl PubMed [122]. Al-bayatı , R.S.A. , 2024 . Detection and measurement of multilevel COVID-19 infection using gamma correction and features extracted by CNN enhanced with xgboost from CT scan images (Master’s thesis, Kirsehir Ahi Evran University (Turkey)). [123]. ↵ Tseng , C.J. and Tang , C ., 2023 . An optimized XGBoost technique for accurate brain tumor detection using feature selection and image segmentation . Healthcare Analytics , 4 , p. 100217 . OpenUrl [124]. ↵ Gupta , S. , Gupta , S. , Agrawal , A. , Naaz , M. , Yadav , R. and Bagade , P ., 2025 . EXACT-CT: EXplainable Analysis for Crohn’s and Tuberculosis using CT . arXiv preprint arXiv:2503.00159 . [125]. Miao , Y. , Tang , S. , Zhang , Z. , Song , J. , Liu , Z. , Chen , Q. and Zhang , M ., 2024 . Application of deep learning and XGBoost in predicting pathological staging of breast cancer MR images . The Journal of Supercomputing , 80 ( 7 ), pp. 8933 – 8953 . OpenUrl [126]. ↵ Khan , W.Z. , Azam , F. and Khan , M.K ., 2022 . Deep-Learning-Based COVID-19 Detection: Challenges and Future Directions . IEEE Transactions on Artificial Intelligence , 4 ( 2 ), pp. 210 – 228 . OpenUrl View the discussion thread. Back to top Previous Next Posted June 30, 2025. Download PDF Data/Code Email Thank you for your interest in spreading the word about medRxiv. NOTE: Your email address is requested solely to identify you as the sender of this article. Your Email * Your Name * Send To * Enter multiple addresses on separate lines or separate them with commas. You are going to email the following Efficient Chest X-Ray Feature Extraction and Feature Fusion for Pneumonia Detection Using Lightweight Pretrained Deep Learning Models Message Subject (Your Name) has forwarded a page to you from medRxiv Message Body (Your Name) thought you would like to see this page from the medRxiv website. Your Personal Message CAPTCHA This question is for testing whether or not you are a human visitor and to prevent automated spam submissions. Share Efficient Chest X-Ray Feature Extraction and Feature Fusion for Pneumonia Detection Using Lightweight Pretrained Deep Learning Models Yashvi Chandola , Vivek Uniyal , Yamini Bachheti medRxiv 2025.06.28.25330459; doi: https://doi.org/10.1101/2025.06.28.25330459 Share This Article: Copy Citation Tools Efficient Chest X-Ray Feature Extraction and Feature Fusion for Pneumonia Detection Using Lightweight Pretrained Deep Learning Models Yashvi Chandola , Vivek Uniyal , Yamini Bachheti medRxiv 2025.06.28.25330459; doi: https://doi.org/10.1101/2025.06.28.25330459 Citation Manager Formats BibTeX Bookends EasyBib EndNote (tagged) EndNote 8 (xml) Medlars Mendeley Papers RefWorks Tagged Ref Manager RIS Zotero Tweet Widget Facebook Like Google Plus One Subject Area Radiology and Imaging Subject Areas All Articles Addiction Medicine (566) Allergy and Immunology (863) Anesthesia (297) Cardiovascular Medicine (4411) Dentistry and Oral Medicine (443) Dermatology (380) Emergency Medicine (606) Endocrinology (including Diabetes Mellitus and Metabolic Disease) (1505) Epidemiology (15205) Forensic Medicine (30) Gastroenterology (1119) Genetic and Genomic Medicine (6575) Geriatric Medicine (666) Health Economics (994) Health Informatics (4511) Health Policy (1365) Health Systems and Quality Improvement (1608) Hematology (537) HIV/AIDS (1263) Infectious Diseases (except HIV/AIDS) (15903) Intensive Care and Critical Care Medicine (1103) Medical Education (620) Medical Ethics (144) Nephrology (666) Neurology (6573) Nursing (345) Nutrition (998) Obstetrics and Gynecology (1139) Occupational and Environmental Health (954) Oncology (3319) Ophthalmology (968) Orthopedics (369) Otolaryngology (420) Pain Medicine (435) Palliative Medicine (129) Pathology (662) Pediatrics (1689) Pharmacology and Therapeutics (691) Primary Care Research (710) Psychiatry and Clinical Psychology (5423) Public and Global Health (9205) Radiology and Imaging (2191) Rehabilitation Medicine and Physical Therapy (1367) Respiratory Medicine (1191) Rheumatology (593) Sexual and Reproductive Health (709) Sports Medicine (529) Surgery (709) Toxicology (99) Transplantation (288) Urology (265) (function(){function c(){var b=a.contentDocument||a.contentWindow.document;if(b){var d=b.createElement('script');d.innerHTML="window.__CF$cv$params={r:'9fec770dbd2758f4',t:'MTc3OTI5MTg5MA=='};var a=document.createElement('script');a.src='/cdn-cgi/challenge-platform/scripts/jsd/main.js';document.getElementsByTagName('head')[0].appendChild(a);";b.getElementsByTagName('head')[0].appendChild(d)}}if(document.body){var a=document.createElement('iframe');a.height=1;a.width=1;a.style.position='absolute';a.style.top=0;a.style.left=0;a.style.border='none';a.style.visibility='hidden';document.body.appendChild(a);if('loading'!==document.readyState)c();else if(window.addEventListener)document.addEventListener('DOMContentLoaded',c);else{var e=document.onreadystatechange||function(){};document.onreadystatechange=function(b){e(b);'loading'!==document.readyState&&(document.onreadystatechange=e,c())}}}})();