A Few-shot Diabetes Foot Ulcer Image Classification Method Based on Deep ResNet and Transfer Learning

A Few-shot Diabetes Foot Ulcer Image Classification Method Based on Deep ResNet and Transfer Learning | 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 Article A Few-shot Diabetes Foot Ulcer Image Classification Method Based on Deep ResNet and Transfer Learning Cheng Wang, Zhen Yu, Zhou Long, Hui Zhao This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-4819913/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 02 Dec, 2024 Read the published version in Scientific Reports → Version 1 posted 10 You are reading this latest preprint version Abstract Background and Objective: Diabetes foot ulcer (DFU) is one of the common complications of diabetes patients, which may lead to infection, necrosis and even amputation. Therefore, early diagnosis, classification of severity and related treatment are crucial for the patients. However, traditional DFU classification methods often require experienced doctors to manually classify infections or non-infections, ischemia or non-ischemia, which is time-consuming and low accuracy. The objective of the study is to propose a few-shot DFU image classification method based on deep residual neural network and transfer learning. Methods Considering the difficulty in obtaining clinical DFU images, it is a few-shot problem. Therefore, the methods include: (1)Data augmentation of the original DFU images by using geometric transformations and random noise; (2)Deep ResNet models selection based on different convolutional layers comparative experiments; (3)DFU classification model training with transfer learning by using the selected pre-trained ResNet model and fine tuning. Results To verify the proposed classification method, the experiments were performed with the original and augmented image datasets by separating three classifications:zero grade, mild grade, severe grade. (1) The datasets were augmented from the original 146 to 3000 image datasets and the corresponding DFU classification’s average accuracy from 0.9167 to 0.9867; (2)Comparative experiments were conducted with ResNet18, ResNet34, ResNet50, ResNet101, ResNet152 by using 146 image datasets, and the average accuracy/loss is 0.686/0.4649, 0.588/0.5424, 0.549/0.6036, 0.510/1.7144, 0.471/0.5462 respectively; (3) Based on the augmented 3000 image datasets, it was achieved 0.9867 average accuracy with the DFU classification model, which was trained by the pre-trained ResNet50 and hyper-parameters. Conclusions The experimental results indicated that the proposed few-shot DFU image classification method based on deep residual neural network and transfer learning got very high accuracy, and it can be applied to screening for the diabetes in rural and undeveloped areas. Physical sciences/Engineering/Biomedical engineering Physical sciences/Mathematics and computing/Computer science Diabetes Foot Ulcer classification method deep learning Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 I. Introduction Diabetes foot ulcer (DFU) is one of the common complications of diabetes patients, which may lead to infection, necrosis and even amputation. [ 1 , 2 , 3 ] Therefore, early diagnosis, classification of severity and related treatment are crucial for the patients. [ 4 ] The classification of severity is a part of diagnosis and the basic step of treatments for DFU. In clinics, current classification method of DFU severity is mainly based on Wagner or Texas [ 5 ] scale measurement by experienced doctors, and it is subjective and requires high professionalism. Recently, many artificial intelligence and machine learning related algorithms had been applied to the classification of DFU. Traditional machine learning based DFU classification methods(SVM,KMA) mostly performed binary classification of ischemia or non-ischemia, infection or non-infection, and achieved high accuracy. [ 6 , 7 , 8 ] However, it has the following disadvantages: manual feature extraction, time-consuming for processing a large number of DFU images, and limited application with coarse (binary) classification. [ 9 , 10 ] Different from traditional machine learning, deep learning can automatically extract features. [ 11 ] At present, deep learning based DFU classification methods (CNN,GoogleNet,VGG16etc) mostly also performed binary classification, [ 12 , 13 , 14 ] and the studies of DFU multi-classifications is rare, and the accuracy is not high. [ 15 ] Considering the difficulty in obtaining clinical DFU images [ 16 ] , it is a few-shot problem. Thus, based on the above study backgrounds of DFU classification, the study aims to propose a three-classification method for few-shot DFU images based on deep ResNet [ 17 ] and transfer learning [ 18 ] , with the ultimate goal of yielding high multi-classification accuracy for DFU images. [ 19 ] The main contributions of this paper are as follows: 1) A DFU images datasets photographed and marked by professional clinicians are provided, including 146 original images and 3000 augmented images. All images are made public on GitHub ( https://github.com/git-yuzhen/DFU-classification ). 2) A classification method combining deep learning and transfer learning is proposed to achieve a new three-classification. The remainder of the article is organized as follows. Materials and Methods provides a brief overview of a three-classification method for few-shot DFU images based on deep ResNet and transfer learning, and related experimental results and analyses are given in Experiments and Results and Discussion , Conclusion summarizes the article. II. Materials and Methods The research framework of the article is show in Fig. 1 . Data obtaining Data obtaining environment: including images of different cases and different stages. The picture was taken by a clinician using Apple 13pro max, and the lens was parallel to the ulcer surface, with a distance of 30–40 cm. The shooting environment is indoors under natural light or in an operating room with sufficient light. A total of 146 original images were taken, and the taken photos were marked and divided into three classifications by doctors with professional clinical qualifications: 27 images for zero grade, 45 images for mild grade and 74 images for severe grade. The principle of the three-classification is based on Wagner classification and DFU severity (illustrated in Fig. 2 ) as it is showed in Table 1 . This study is a retrospective study, following the ethical guidelines of Helsinki Declaration and approved by the Ethics Committee of Beijing Chaoyang Hospital affiliated to Capital Medical University (ethical batch number: 2022-ke-629). All patients provided written informed consent before operation. Table 1 Wagner classification and new class Wagner Grade 0 Grade 1 Grade 2 Grade 3 Grade 4 Grade 5 New class zero grade mild grade severe grade Data augmentation In order to improve the generalization ability of the DFU classification model, the original datasets were augmented by enlarging and reducing the size [ 20 ] , rotating (90°), flipping (horizontal flipping, vertical flipping), brightness changing (brightening, darkening), pixel shifting [ 21 , 22 ] , adding noise (salt and pepper noise, gaussian noise [ 23 , 24 ] ) and so on. The datasets were divided into training set, test set and verification set for training and evaluation of the model. Images of the datasets after data augmentation are illustrated in Fig. 3 . Deep ResNet models selection Resnet series models were selected as the baseline models: ResNet18, ResNet34, ResNet50, ResNet101 and ResNet152 were tested respectively, and the model with the best experimental results was selected as the pre-trained model. The model was trained on the training set, and the cross-entropy error was defined as the loss function. $$\:L=-\frac{1}{m}\sum\:_{i=1}^{m}{\sum\:}_{k=1}^{n}{y}_{\:\:k}^{\left(i\right)}\text{l}\text{o}\text{g}\left({p}_{\:\:k}^{\left(i\right)}\right)$$ 1 L represents the cross-entropy error, defining m samples, n categories, \(\:{y}_{\:\:k}^{\left(i\right)}\) represents the true labeling of the k-th category for the i-th sample, and \(\:{p}_{\:\:k}^{\left(i\right)}\) represents the predicted probability of the model for the k-th category for the i-th sample. The stochastic gradient descent method was defined as the optimizer. $$\:\left\{\begin{array}{c}g=\frac{1}{m}{\nabla\:}_{\theta\:}{\sum\:}_{i=0}^{m}\text{L}(f\left({x}_{i,}\theta\:\right),{y}_{i})\\\:\theta\:\leftarrow\:\theta\:-\epsilon\:g\end{array}\right.$$ 2 g represents the gradient, ∇θL(xi,θ) represents the gradient of the loss function for all samples, m samples are randomly selected from within n samples at a time, ε is the learning rate. During the training process, we monitor the performance of the model on the verification set to avoid over-fitting. Evaluation indicators include average accuracy and average loss. In this paper, the average accuracy is defined as this formula: $$\:\overline{A}=\frac{1}{n}{\sum\:}_{i=1}^{i=n}{x}_{i}$$ 3 \(\:\overline{A}\) represents the average accuracy, n represents the total number of rounds of training, and \(\:{x}_{i}\) represents the accuracy of the i-th round of training. The average loss is defined as this formula: $$\:\overline{L}=\frac{1}{n}{\sum\:}_{i=1}^{i=n}{y}_{i}$$ 4 \(\:\overline{L}\) represents the average loss, n represents the total number of rounds of training, and \(\:{y}_{i}\) represents the loss of the i-th round of training. According to the above method, the detailed model selection comparison experiments and hyper-parameters setting were conducted in Experiment 2 , and ResNet50 was selected. DFU classification model training DFU classification model training method is using the pre-trained model and hyper-parameters fine-tuning, and ResNet50 is the pre-trained model. In deep learning, the selection of hyper-parameters usually includes: image_size, learning rate, learning rate decay, batch_size, number of rounds(epoch). [ 25 ] Based on the accuracy and loss of different ResNet models, the most suitable ResNet model for performing dataset B (Table 3 ) is analyzed as the final classification model. III. Experiments and Results Windows 10 was used as the operating system for the experiment, PyTorch version 2.2.0 was used as the deep learning development framework, and Python version 3.8 was used as the development language. The CPU used in the experiment was Intel Core i5-8250U, and the GPU was NVIDIA GeForce RTX 3060. The detailed experimental environment is showed in Table 2 . Table 2 Experimental environment Operating System Windows 10 Python 3.8 PyTorch 2.2.0 Cuda 10.1 CPU Intel Core i5-8250U GPU NVIDIA GeForce RTX 3060 In this section, three experiments were conducted. Experiment 1 was to verify the effect of DFU data augmentation. Experiment 2 is to select pre-trained model. Experiment 3 is to verify the effectiveness of DFU classification model. And the experimental related data are showed in Table 3 . Table 3 Experimental data of DFU DFU datasets A * train test valid zero grade 27 19 3 5 mild grade 45 31 5 9 severe grade 74 52 7 15 total 146 102 15 29 DFU datasets B * train test valid zero grade 1000 700 100 200 mild grade 1000 700 100 200 severe grade 1000 700 100 200 total 3000 2100 300 600 * DFU datasets A is the original datasets, and DFU datasets B is the augmented datasets. Experiment 1 After data augmentation, the datasets A had been increased from 146 (including 27 for zero grade, 45 for mild grade and 74 for severe grade) to 3,000 (including 1000 for zero grade, 1000 for mild grade and 1000 for severe grade). And the average accuracy of DFU classification reached 0.9867 from 0.9167. Experiment 2 Cuda is used to train the datasets A with different ResNet models. The cross entropy error was defined as the loss function. The stochastic gradient descent method was defined as the optimizer. The learning rate was set to 0.001. The random seed was 42. The last fully connected layer of the models was modified to match the number of categories of the datasets and make three classifications. Compared with ResNet18, ResNet34, ResNet50, ResNet101 and ResNet152, datasets A were used with a total of 10 rounds training, and the average accuracy/loss were 0.686/0.649, 0.588/0.524, 0.549/0.6036, 0.510/1.7144 and 0.44 respectively. The experimental results are showed in Table 4 and Fig. 4 . Experiment 3 : The datasets A and the datasets B were trained respectively using the selected ResNet50 as a pre-trained model. Record the experimental results, including average loss, average accuracy. The training hyper-parameters of ResNet50 for the datasets A are: img_size = 224, learning rate = 0.001, learning rate decay = 0.00001, batch_size = 6, epochs = 20. The average accuracy of the experiment was 0.9167 and the average loss was 0.2546. The training results of the datasets A are showed in Fig. 5 . The training hyper-parameters of ResNet50 with the datasets B are: img_size = 224, learning rate = 0.001, learning rate decay = 0.00001, batch_size = 6, epochs = 5. The average accuracy of the experiment was 0.9867 and the average loss was 0.0398. The training results of the datasets B are showed in Fig. 6 . Table 4 Comparative experiment of different residual models in the datasets A. Model ResNet18 ResNet34 ResNet50 ResNet101 ResNet152 Accuracy 68.6% 58.8% 54.9% 51.0% 47.1% Avg loss 0.4649 0.5424 0.6036 1.7144 0.5462 IV. Discussion The images were shot by professional clinicians in a standardized way, and 146 original data sets (datasets A) were obtained, of which 27 were zero grade, 45 were mild grade and 74 were severe grade. The shooting location is usually collected in an outpatient clinic with sufficient natural light or an operating room with sufficient light. The datasets A are augmented by using geometric transformation and random noise, and the amplified datasets reach 3000, including 1000 for zero grade, 1000 for mild grade and 1000 for severe grade. After data augmentation, the uneven image classification is balanced. The paper proposed a few-shot DFU image classification method based on deep residual neural network and transfer learning, and related experiments are given. From the experimental results of Experiment 2 , it can be found that when using small datasets, with the increase of network depth, the accuracy of the experiment is gradually decreasing. This is because deeper networks usually have stronger expressive ability and can capture more complex features and patterns, which may perform better in some complex tasks. However, in some simple classification tasks, deeper network may lead to over-fitting or training difficulties, thus reducing the accuracy. When the amount of data is only146, ResNet18 and ResNet34 converge more easily and show higher accuracy, and perform better on datasets because they have appropriate hyper-parameters to deal with the characteristics of tasks, and are less prone to over-fitting. So when the amount of data increases, the corresponding network structure should increase. Considering the influence of training time and computing power, according to the size of the datasets B after data augmentation, we choose the median value and take ResNet50 as the pre-trained model in order to get better training effect. Experimental results of Experiment 1&3 show that the proposed method has achieved excellent performance in the image classification task of DFU, and the augmented data obviously helped to improve the experimental accuracy. After data augmentation, the accuracy of the model on the verification set reached 0.9867 from 0.9167. It showed that the proposed method can accurately classify DFU images. However, images are collected in a standardized environment, and the influence of other complex environments on the accuracy of image classification is not considered. V. Conclusion The experimental results show that the proposed method has achieved satisfactory performance in the image classification task of DFU, and it is suitable for the general screening of diabetic patients in rural and underdeveloped areas, so as to reduce the cost of their long journey to hospitals or health centers. This study provides new possibilities for the diagnosis and treatment of diabetic patients and is expected to play an important role in clinical practice. Declarations Data Availability Statement The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation. Any data or questions regarding the study, please kindly contact Zhen Yu( [email protected] ) and Cheng Wang (corresponding author, [email protected] ). Ethics Statement The study was reviewed and approved by the institutional review board of Beijing Chaoyang Hospital affiliated to Capital Medical University (number is 2022-ke-629). All participants have signed informed consent forms. Author Contributions Zhen Yu did the experiments. Zhen Yu, Zhou Long and Cheng Wang analyzed experimental data. Cheng Wang and Hui Zhao designed the research. Zhen Yu and Cheng Wang together as co-author wrote the manuscript, and both are the first authors of the manuscript. All authors contributed to the article and approved the submitted version. Funding This work is partly supported by Shandong Province Science and Technology Small and Medium sized Enterprise Innovation Capability Enhancement Project (2022TSGC2532), 2025 Key Tech-nology Innovation Program of Ningbo(2020Z014) and Capital Health Development Research Special Project (2024-2-7042). The authors are grateful for these supports. Conflict of Interest The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. References Jiang Y, Wang X, Lei X, et al. A cohort study of diabetic patients and diabetic footulceration patients in China[J]. Wound Repair and Regeneration, 2015,23(2). Jiang Qixia, Geng Guangli, Chang Yan, et al. Prevention of foot ulcer in 188 diabetic patients [J]. Chinese Journal of Nursing, 2001(2):85-87. Xu Y, Wang L, He J, et al. Prevalence and Control of Diabetes in Chinese Adults.JAMA. 2013;310(9):948–959. doi:10.1001/jama.2013.168118 Wang Aihong, Zhao Kun, Li Qiang, et al. Investigation and medical economic analysis of diabetic foot in some provinces and cities of China [J]. chinese journal of endocrinology and metabolism, 2005, 21(6): 496-499. Oyibo S O, Jude E B, Tarawneh I, et al. A comparison of two diabetic foot ulcer classification systems: the Wagner and the University of Texas wound classification systems[J]. Diabetes care, 2001, 24(1): 84-88. Wang L, Pedersen PC, Agu E, Strong DM, Tulu B. Area determination of diabetic foot ulcer images using a cascaded two-stage SVM-based classification. IEEE Trans Biomed Eng (2016) 64(9):2098–109. doi: 10.1109/TBME.2016.2632522 Patel S, Patel R, Desai D. Diabetic foot ulcer wound tissue detection and classification[C]//2017 international conference on innovations in information, embedded andcommunication systems (ICIIECS). IEEE, 2017: 1-5. Yadav MK, Manohar DD, Mukherjee G, Chakraborty C. Segmentation of Chronic Wound Areas by Clustering Techniques Using Selected Color Space[J]. Journal of Medical Imaging and Health Informatics, 2013,3(1): 22-29. Goyal M, Reeves N D, Rajbhandari S, et al. Robust methods for real-time diabeticfoot ulcer detection and localization on mobile devices[J]. IEEE journal of biomedical and health informatics, 2018, 23(4): 1730-1741. Chan H P, Samala R K, Hadjiiski L M, et al. Deep learning in medical image analysis[J]. Deep learning in medical image analysis: challenges and applications, 2020:3-21. Min S, Lee B, Yoon S. Deep learning in bioinformatics[J]. Briefings in bioinformatics, 2017, 18(5): 851-869. Goyal M, Reeves ND, Davison AK, Rajbhandari S, Spragg J, Yap MH. Dfunet: Convolutional neural networks for diabetic foot ulcer classification. IEEE Trans Emerg Topic Comput Intell (2018) 4(5):728–39. doi: 10.1109/TETCI.2018. 2866254 Alzubaidi L, Fadhel MA, Oleiwi SR, Al-Shamma O, Zhang J. DFU_QUTNet: diabetic foot ulcer classification using novel deep convolutional neural network. Multimed Tools Appl (2020) 79(21):15655–77. doi: 10.1007/ s11042-019-07820-w Xu Y, Han K, Zhou Y, Wu J, Xie X, Xiang W. Classification of diabetic foot ulcers using class knowledge banks. Front Bioengineer Biotechnol (2021) 9. doi: 10.3389/fbioe.2021.811028 Cruz-Vega I, Hernandez-Contreras D, Peregrina-Barreto H, RangelMagdaleno JJ, Ramirez-Cortes JM. Deep learning classification for diabetic foot thermograms. Sensors (2020) 20(6):1762. doi: 10.3390/s20061762 Yap M H, Kendrick C, Reeves N D, et al. Development of diabetic foot ulcer datasets: an overview[J]. Diabetic Foot Ulcers Grand Challenge, 2021: 1-18. He K, Zhang X, Ren S, et al. Deep residual learning for image recognition[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2016: 770-778. Vaswani A, Shazeer N, Parmar N, et al. Attention is all you need[J]. Advances in neural information processing systems, 2017, 30. Ahsan M, Naz S, Ahmad R, et al. A deep learning approach for diabetic foot ulcerclassification and recognition[J]. Information, 2023, 14(1): 36. Devries, T., Taylor, G.W.: Improved Regularization of Convolutional Neural Net-works with Cutout. 1708.04552 (2017) Zhong, Z., Zheng, L., Kang, G., Li, S., Yang, Y.: Random Erasing Data Augmentation. In: The Thirty-Fourth AAAI Conference on Artifificial Intelligence, AAAI 2020, The Thirty- Second Innovative Applications of Artifificial Intelligence Conference, IAAI 2020, The Tenth AAAI Symposium on Educational Advances in Artifificial Intelligence, EAAI 2020, New York, NY, USA, February 7-12, 2020. pp. 13001–13008. AAAI Press (2020) Takahashi, R., Matsubara, T., Uehara, K.: RICAP: Random Image Cropping and Patching Data Augmentation for Deep CNNs. In: Zhu, J., Takeuchi, I. (eds.) Proceedings of The 10th Asian Conference on Machine Learning, ACML 2018, Beijing, China, November 14-16, 2018. Proceedings of Machine Learning Research, vol. 95, pp. 786–798. PMLR (2018) Lopes, R.G., Yin, D., Poole, B., Gilmer, J., Cubuk, E.D.: Improving Robustness Without Sacrifificing Accuracy with Patch Gaussian Augmentation. 1906.02611 (2019) Zhou, K., Yang, Y., Qiao, Y., Xiang, T.: Domain generalization with mixstyle. CoRR 2104.02008 (2021) Bengio Y. Practical recommendations for gradient-based training of deep architectures[M]//Neural networks: Tricks of the trade: Second edition. Berlin, Heidelberg: Springer Berlin Heidelberg, 2012: 437-478. Additional Declarations No competing interests reported. Cite Share Download PDF Status: Published Journal Publication published 02 Dec, 2024 Read the published version in Scientific Reports → Version 1 posted Editorial decision: Revision requested 11 Oct, 2024 Reviews received at journal 08 Oct, 2024 Reviews received at journal 30 Aug, 2024 Reviewers agreed at journal 27 Aug, 2024 Reviewers agreed at journal 27 Aug, 2024 Reviewers invited by journal 27 Aug, 2024 Editor assigned by journal 20 Aug, 2024 Editor invited by journal 08 Aug, 2024 Submission checks completed at journal 07 Aug, 2024 First submitted to journal 29 Jul, 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-4819913","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":347879505,"identity":"bb22f2ff-3ccf-42fe-97e7-aeb87f7e5d35","order_by":0,"name":"Cheng Wang","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA7UlEQVRIiWNgGAWjYLACxgYgwczY+CCBNC3szc0GCQwGpGjhOd4mwUCMFoPjZw+/+LnDJk8+IrGt4kHNHwb+9gOMnwvwaTmTl2bZeyat2PBGYtuNhGMGDBJnEpilZ+DRYnYgx8yYse1w4sYZQC2JDUCH3WBgY+bBp+X8G5CW/2AtBSAt8gS13MgxfszYdiBxPs/BNgaQFgNCWuxvvDFj7G1LTtzA3tgskXDMmMfwTGKzND4tkv05xh9+ttklzm9mf/jxR42cnNzxwwc/49MCBGwSINLgAITHA40mvID5A4iUJ6huFIyCUTAKRiwAAJX+UHRPV7L5AAAAAElFTkSuQmCC","orcid":"","institution":"Shandong Academy of Intelligent Computing Technology, Shandong Institutes of Industrial Technology(SDIIT)","correspondingAuthor":true,"prefix":"","firstName":"Cheng","middleName":"","lastName":"Wang","suffix":""},{"id":347879507,"identity":"c8f41e2c-d2f9-47cb-9a7b-c7fbe256152f","order_by":1,"name":"Zhen Yu","email":"","orcid":"","institution":"Qilu University of Technology (Shandong Academy of Sciences)","correspondingAuthor":false,"prefix":"","firstName":"Zhen","middleName":"","lastName":"Yu","suffix":""},{"id":347879508,"identity":"82889e48-7ba5-4e07-b027-1165cf74b29f","order_by":2,"name":"Zhou Long","email":"","orcid":"","institution":"Bejing Key Laboratory of Mobile Computing and Pervasive Device, Institute of Computing Technology (ICT)","correspondingAuthor":false,"prefix":"","firstName":"Zhou","middleName":"","lastName":"Long","suffix":""},{"id":347879511,"identity":"a9eca794-c661-4db8-9c25-2e076368c178","order_by":3,"name":"Hui Zhao","email":"","orcid":"","institution":"Orthopedics Department, Beijing Chao-Yang Hospital, Capital Medical University","correspondingAuthor":false,"prefix":"","firstName":"Hui","middleName":"","lastName":"Zhao","suffix":""}],"badges":[],"createdAt":"2024-07-29 07:31:19","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-4819913/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-4819913/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1038/s41598-024-80691-w","type":"published","date":"2024-12-02T15:57:58+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":64568623,"identity":"57373b9e-e070-4a4f-a4f4-11bc86266471","added_by":"auto","created_at":"2024-09-16 00:41:53","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":53410,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eA Few-shot DFU Image Classification Method Based on Deep ResNet and Transfer Learning\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-4819913/v1/068a6703285aa575fe765571.png"},{"id":64569354,"identity":"4d28f827-52f3-4fc4-9be1-803559028612","added_by":"auto","created_at":"2024-09-16 00:49:53","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":223377,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eNew class (a) zero grade (b) mild grade (c) severe grade\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-4819913/v1/601a76b3e62e3e7d809ec196.png"},{"id":64568625,"identity":"254bbf23-a3ca-49bc-81ab-0c2aeb201240","added_by":"auto","created_at":"2024-09-16 00:41:53","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":411599,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eDFU images augmentation mode (a) original image (b) rotated90° (c) flipped horizontally (d) flipped vertically (e) moved (f) gaussian noise.\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-4819913/v1/9dc875f71a1f17a0a59f1e18.png"},{"id":64568622,"identity":"bcffab40-77bb-4320-96fb-3bbd95ba7897","added_by":"auto","created_at":"2024-09-16 00:41:53","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":30088,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eComparative experiment of different residual models in the datasets A.\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-4819913/v1/99e09ab4e5cb71fc8f707dd0.png"},{"id":64568626,"identity":"94affc90-7e54-4a38-ab02-067548c76ee0","added_by":"auto","created_at":"2024-09-16 00:41:53","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":224159,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003ePre-trained, datasets A.\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"5.png","url":"https://assets-eu.researchsquare.com/files/rs-4819913/v1/32e386412a296bed88a7b8b4.png"},{"id":64569355,"identity":"c6d41187-bc45-49c5-9817-00e156a74163","added_by":"auto","created_at":"2024-09-16 00:49:53","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":75358,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003ePre-trained, datasets B.\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"6.png","url":"https://assets-eu.researchsquare.com/files/rs-4819913/v1/bfbb3eb689a08f24e7e40099.png"},{"id":70964770,"identity":"e6fba7a4-59ee-4136-8934-7b168c15ca9b","added_by":"auto","created_at":"2024-12-09 16:15:43","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1765712,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-4819913/v1/8acaeeb8-99bc-4674-b288-4de4484f7b1d.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"A Few-shot Diabetes Foot Ulcer Image Classification Method Based on Deep ResNet and Transfer Learning","fulltext":[{"header":"I. Introduction","content":"\u003cp\u003eDiabetes foot ulcer (DFU) is one of the common complications of diabetes patients, which may lead to infection, necrosis and even amputation.\u003csup\u003e[\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e, \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e, \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]\u003c/sup\u003e Therefore, early diagnosis, classification of severity and related treatment are crucial for the patients.\u003csup\u003e[\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]\u003c/sup\u003e The classification of severity is a part of diagnosis and the basic step of treatments for DFU. In clinics, current classification method of DFU severity is mainly based on Wagner or Texas\u003csup\u003e[\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]\u003c/sup\u003e scale measurement by experienced doctors, and it is subjective and requires high professionalism. Recently, many artificial intelligence and machine learning related algorithms had been applied to the classification of DFU. Traditional machine learning based DFU classification methods(SVM,KMA) mostly performed binary classification of ischemia or non-ischemia, infection or non-infection, and achieved high accuracy.\u003csup\u003e[\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e, \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e, \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]\u003c/sup\u003e However, it has the following disadvantages: manual feature extraction, time-consuming for processing a large number of DFU images, and limited application with coarse (binary) classification.\u003csup\u003e[\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e, \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]\u003c/sup\u003e Different from traditional machine learning, deep learning can automatically extract features.\u003csup\u003e[\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]\u003c/sup\u003e At present, deep learning based DFU classification methods (CNN,GoogleNet,VGG16etc) mostly also performed binary classification,\u003csup\u003e[\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e, \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]\u003c/sup\u003e and the studies of DFU multi-classifications is rare, and the accuracy is not high.\u003csup\u003e[\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]\u003c/sup\u003e\u003c/p\u003e \u003cp\u003eConsidering the difficulty in obtaining clinical DFU images\u003csup\u003e[\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e]\u003c/sup\u003e, it is a few-shot problem. Thus, based on the above study backgrounds of DFU classification, the study aims to propose a three-classification method for few-shot DFU images based on deep ResNet\u003csup\u003e[\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]\u003c/sup\u003e and transfer learning\u003csup\u003e[\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]\u003c/sup\u003e, with the ultimate goal of yielding high multi-classification accuracy for DFU images. \u003csup\u003e[\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]\u003c/sup\u003e The main contributions of this paper are as follows: 1) A DFU images datasets photographed and marked by professional clinicians are provided, including 146 original images and 3000 augmented images. All images are made public on GitHub (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://github.com/git-yuzhen/DFU-classification\u003c/span\u003e\u003cspan address=\"https://github.com/git-yuzhen/DFU-classification\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e). 2) A classification method combining deep learning and transfer learning is proposed to achieve a new three-classification.\u003c/p\u003e \u003cp\u003eThe remainder of the article is organized as follows. \u003cem\u003eMaterials and Methods\u003c/em\u003e provides a brief overview of a three-classification method for few-shot DFU images based on deep ResNet and transfer learning, and related experimental results and analyses are given in \u003cem\u003eExperiments and Results\u003c/em\u003e and \u003cem\u003eDiscussion\u003c/em\u003e, \u003cem\u003eConclusion\u003c/em\u003e summarizes the article.\u003c/p\u003e"},{"header":"II. Materials and Methods","content":"\u003cp\u003eThe research framework of the article is show in Fig. \u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e\n\u003cp\u003eData obtaining\u003c/p\u003e\n\u003cp\u003eData obtaining environment: including images of different cases and different stages. The picture was taken by a clinician using Apple 13pro max, and the lens was parallel to the ulcer surface, with a distance of 30\u0026ndash;40 cm. The shooting environment is indoors under natural light or in an operating room with sufficient light. A total of 146 original images were taken, and the taken photos were marked and divided into three classifications by doctors with professional clinical qualifications: 27 images for zero grade, 45 images for mild grade and 74 images for severe grade. The principle of the three-classification is based on Wagner classification and DFU severity (illustrated in Fig. \u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e) as it is showed in Table \u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e\n\u003cp\u003eThis study is a retrospective study, following the ethical guidelines of Helsinki Declaration and approved by the Ethics Committee of Beijing Chaoyang Hospital affiliated to Capital Medical University (ethical batch number: 2022-ke-629). All patients provided written informed consent before operation.\u003c/p\u003e\u003ctable id=\"Tab1\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eWagner classification and new class\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003ccolgroup cols=\"7\"\u003e\u003c/colgroup\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eWagner\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eGrade 0\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eGrade 1\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eGrade 2\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eGrade 3\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eGrade 4\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eGrade 5\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eNew class\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ezero grade\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003emild grade\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"3\"\u003e\n \u003cp\u003esevere grade\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003eData augmentation\u003c/p\u003e\n\u003cp\u003eIn order to improve the generalization ability of the DFU classification model, the original datasets were augmented by enlarging and reducing the size\u003csup\u003e[\u003cspan class=\"CitationRef\"\u003e20\u003c/span\u003e]\u003c/sup\u003e, rotating (90\u0026deg;), flipping (horizontal flipping, vertical flipping), brightness changing (brightening, darkening), pixel shifting\u003csup\u003e[\u003cspan class=\"CitationRef\"\u003e21\u003c/span\u003e, \u003cspan class=\"CitationRef\"\u003e22\u003c/span\u003e]\u003c/sup\u003e, adding noise (salt and pepper noise, gaussian noise\u003csup\u003e[\u003cspan class=\"CitationRef\"\u003e23\u003c/span\u003e, \u003cspan class=\"CitationRef\"\u003e24\u003c/span\u003e]\u003c/sup\u003e) and so on. The datasets were divided into training set, test set and verification set for training and evaluation of the model. Images of the datasets after data augmentation are illustrated in Fig. \u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003e.\u003c/p\u003e\n\u003cp\u003eDeep ResNet models selection\u003c/p\u003e\n\u003cp\u003eResnet series models were selected as the baseline models: ResNet18, ResNet34, ResNet50, ResNet101 and ResNet152 were tested respectively, and the model with the best experimental results was selected as the pre-trained model. The model was trained on the training set, and the cross-entropy error was defined as the loss function.\u003c/p\u003e\n\u003cdiv id=\"Equ1\" class=\"Equation\"\u003e\n \u003cdiv class=\"mathdisplay\" id=\"FileID_Equ1\" name=\"EquationSource\"\u003e$$\\:L=-\\frac{1}{m}\\sum\\:_{i=1}^{m}{\\sum\\:}_{k=1}^{n}{y}_{\\:\\:k}^{\\left(i\\right)}\\text{l}\\text{o}\\text{g}\\left({p}_{\\:\\:k}^{\\left(i\\right)}\\right)$$\u003c/div\u003e\n \u003cdiv class=\"EquationNumber\"\u003e1\u003c/div\u003e\n\u003c/div\u003e\n\u003cp\u003eL represents the cross-entropy error, defining m samples, n categories, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{y}_{\\:\\:k}^{\\left(i\\right)}\\)\u003c/span\u003e\u003c/span\u003e represents the true labeling of the k-th category for the i-th sample, and \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{p}_{\\:\\:k}^{\\left(i\\right)}\\)\u003c/span\u003e\u003c/span\u003e represents the predicted probability of the model for the k-th category for the i-th sample.\u003c/p\u003e\n\u003cp\u003eThe stochastic gradient descent method was defined as the optimizer.\u003c/p\u003e\n\u003cdiv id=\"Equ2\" class=\"Equation\"\u003e\n \u003cdiv class=\"mathdisplay\" id=\"FileID_Equ2\" name=\"EquationSource\"\u003e$$\\:\\left\\{\\begin{array}{c}g=\\frac{1}{m}{\\nabla\\:}_{\\theta\\:}{\\sum\\:}_{i=0}^{m}\\text{L}(f\\left({x}_{i,}\\theta\\:\\right),{y}_{i})\\\\\\:\\theta\\:\\leftarrow\\:\\theta\\:-\\epsilon\\:g\\end{array}\\right.$$\u003c/div\u003e\n \u003cdiv class=\"EquationNumber\"\u003e2\u003c/div\u003e\n\u003c/div\u003e\n\u003cp\u003eg represents the gradient, \u0026nabla;\u0026theta;L(xi,\u0026theta;) represents the gradient of the loss function for all samples, m samples are randomly selected from within n samples at a time, \u0026epsilon; is the learning rate.\u003c/p\u003e\n\u003cp\u003eDuring the training process, we monitor the performance of the model on the verification set to avoid over-fitting. Evaluation indicators include average accuracy and average loss. In this paper, the average accuracy is defined as this formula:\u003c/p\u003e\n\u003cdiv id=\"Equ3\" class=\"Equation\"\u003e\n \u003cdiv class=\"mathdisplay\" id=\"FileID_Equ3\" name=\"EquationSource\"\u003e$$\\:\\overline{A}=\\frac{1}{n}{\\sum\\:}_{i=1}^{i=n}{x}_{i}$$\u003c/div\u003e\n \u003cdiv class=\"EquationNumber\"\u003e3\u003c/div\u003e\n\u003c/div\u003e\n\u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u0026nbsp;\u003cspan class=\"mathinline\"\u003e\\(\\:\\overline{A}\\)\u003c/span\u003e\u0026nbsp;\u003c/span\u003e represents the average accuracy, n represents the total number of rounds of training, and \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{x}_{i}\\)\u003c/span\u003e\u003c/span\u003e represents the accuracy of the i-th round of training.\u003c/p\u003e\n\u003cp\u003eThe average loss is defined as this formula:\u003c/p\u003e\n\u003cdiv id=\"Equ4\" class=\"Equation\"\u003e\n \u003cdiv class=\"mathdisplay\" id=\"FileID_Equ4\" name=\"EquationSource\"\u003e$$\\:\\overline{L}=\\frac{1}{n}{\\sum\\:}_{i=1}^{i=n}{y}_{i}$$\u003c/div\u003e\n \u003cdiv class=\"EquationNumber\"\u003e4\u003c/div\u003e\n\u003c/div\u003e\n\u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u0026nbsp;\u003cspan class=\"mathinline\"\u003e\\(\\:\\overline{L}\\)\u003c/span\u003e\u0026nbsp;\u003c/span\u003e represents the average loss, n represents the total number of rounds of training, and \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{y}_{i}\\)\u003c/span\u003e\u003c/span\u003e represents the loss of the i-th round of training.\u003c/p\u003e\n\u003cp\u003eAccording to the above method, the detailed model selection comparison experiments and hyper-parameters setting were conducted in \u003cem\u003eExperiment 2\u003c/em\u003e, and ResNet50 was selected.\u003c/p\u003e\n\u003cp\u003eDFU classification model training\u003c/p\u003e\n\u003cp\u003eDFU classification model training method is using the pre-trained model and hyper-parameters fine-tuning, and ResNet50 is the pre-trained model. In deep learning, the selection of hyper-parameters usually includes: image_size, learning rate, learning rate decay, batch_size, number of rounds(epoch).\u003csup\u003e[\u003cspan class=\"CitationRef\"\u003e25\u003c/span\u003e]\u003c/sup\u003e Based on the accuracy and loss of different ResNet models, the most suitable ResNet model for performing dataset B (Table \u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003e) is analyzed as the final classification model.\u003c/p\u003e"},{"header":"III. Experiments and Results","content":"\u003cp\u003eWindows 10 was used as the operating system for the experiment, PyTorch version 2.2.0 was used as the deep learning development framework, and Python version 3.8 was used as the development language. The CPU used in the experiment was Intel Core i5-8250U, and the GPU was NVIDIA GeForce RTX 3060. The detailed experimental environment is showed in Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e.\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\u003eExperimental environment\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\u003eOperating System\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eWindows 10\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePython\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003e3.8\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePyTorch\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003e2.2.0\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eCuda\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003e10.1\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eCPU\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eIntel Core i5-8250U\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eGPU\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eNVIDIA GeForce RTX 3060\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eIn this section, three experiments were conducted. \u003cem\u003eExperiment 1\u003c/em\u003e was to verify the effect of DFU data augmentation. \u003cem\u003eExperiment 2\u003c/em\u003e is to select pre-trained model. \u003cem\u003eExperiment 3\u003c/em\u003e is to verify the effectiveness of DFU classification model. And the experimental related data are showed in Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e.\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\u003eExperimental data of DFU\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=\"left\" 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\u003eDFU datasets\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eA\u003csup\u003e*\u003c/sup\u003e\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003etrain\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003etest\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003evalid\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ezero grade\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e27\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e19\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e5\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003emild grade\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e45\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e31\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e9\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003esevere grade\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e74\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e52\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e7\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e15\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003etotal\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e146\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e102\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e15\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e29\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eDFU datasets\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eB\u003c/b\u003e\u003csup\u003e\u003cb\u003e*\u003c/b\u003e\u003c/sup\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u003cb\u003etrain\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u003cb\u003etest\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u003cb\u003evalid\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ezero grade\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1000\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e700\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e100\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e200\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003emild grade\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1000\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e700\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e100\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e200\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003esevere grade\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1000\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e700\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e100\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e200\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003etotal\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e3000\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2100\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e300\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e600\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e* \u003cb\u003eDFU datasets A is the original datasets, and DFU datasets B is the augmented datasets.\u003c/b\u003e\u003c/p\u003e \u003cp\u003e \u003cstrong\u003eExperiment 1\u003c/strong\u003e \u003cp\u003eAfter data augmentation, the datasets A had been increased from 146 (including 27 for zero grade, 45 for mild grade and 74 for severe grade) to 3,000 (including 1000 for zero grade, 1000 for mild grade and 1000 for severe grade). And the average accuracy of DFU classification reached 0.9867 from 0.9167.\u003c/p\u003e \u003c/p\u003e \u003cp\u003e \u003cstrong\u003eExperiment 2\u003c/strong\u003e \u003cp\u003eCuda is used to train the datasets A with different ResNet models. The cross entropy error was defined as the loss function. The stochastic gradient descent method was defined as the optimizer. The learning rate was set to 0.001. The random seed was 42. The last fully connected layer of the models was modified to match the number of categories of the datasets and make three classifications. Compared with ResNet18, ResNet34, ResNet50, ResNet101 and ResNet152, datasets A were used with a total of 10 rounds training, and the average accuracy/loss were 0.686/0.649, 0.588/0.524, 0.549/0.6036, 0.510/1.7144 and 0.44 respectively. The experimental results are showed in Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e and Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e.\u003c/p\u003e \u003c/p\u003e \u003cp\u003e \u003cem\u003eExperiment 3\u003c/em\u003e: The datasets A and the datasets B were trained respectively using the selected ResNet50 as a pre-trained model. Record the experimental results, including average loss, average accuracy. The training hyper-parameters of ResNet50 for the datasets A are: img_size\u0026thinsp;=\u0026thinsp;224, learning rate\u0026thinsp;=\u0026thinsp;0.001, learning rate decay\u0026thinsp;=\u0026thinsp;0.00001, batch_size\u0026thinsp;=\u0026thinsp;6, epochs\u0026thinsp;=\u0026thinsp;20. The average accuracy of the experiment was 0.9167 and the average loss was 0.2546. The training results of the datasets A are showed in Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e. The training hyper-parameters of ResNet50 with the datasets B are: img_size\u0026thinsp;=\u0026thinsp;224, learning rate\u0026thinsp;=\u0026thinsp;0.001, learning rate decay\u0026thinsp;=\u0026thinsp;0.00001, batch_size\u0026thinsp;=\u0026thinsp;6, epochs\u0026thinsp;=\u0026thinsp;5. The average accuracy of the experiment was 0.9867 and the average loss was 0.0398. The training results of the datasets B are showed in Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab4\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eComparative experiment of different residual models in the datasets A.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"6\"\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 \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 \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\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\u003eResNet18\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eResNet34\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eResNet50\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eResNet101\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eResNet152\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eAccuracy\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003e68.6%\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u003cb\u003e58.8%\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u003cb\u003e54.9%\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u003cb\u003e51.0%\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u003cb\u003e47.1%\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eAvg loss\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003e0.4649\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u003cb\u003e0.5424\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u003cb\u003e0.6036\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u003cb\u003e1.7144\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u003cb\u003e0.5462\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e"},{"header":"IV. Discussion","content":"\u003cp\u003eThe images were shot by professional clinicians in a standardized way, and 146 original data sets (datasets A) were obtained, of which 27 were zero grade, 45 were mild grade and 74 were severe grade. The shooting location is usually collected in an outpatient clinic with sufficient natural light or an operating room with sufficient light. The datasets A are augmented by using geometric transformation and random noise, and the amplified datasets reach 3000, including 1000 for zero grade, 1000 for mild grade and 1000 for severe grade. After data augmentation, the uneven image classification is balanced. The paper proposed a few-shot DFU image classification method based on deep residual neural network and transfer learning, and related experiments are given.\u003c/p\u003e \u003cp\u003eFrom the experimental results of \u003cem\u003eExperiment 2\u003c/em\u003e, it can be found that when using small datasets, with the increase of network depth, the accuracy of the experiment is gradually decreasing. This is because deeper networks usually have stronger expressive ability and can capture more complex features and patterns, which may perform better in some complex tasks. However, in some simple classification tasks, deeper network may lead to over-fitting or training difficulties, thus reducing the accuracy. When the amount of data is only146, ResNet18 and ResNet34 converge more easily and show higher accuracy, and perform better on datasets because they have appropriate hyper-parameters to deal with the characteristics of tasks, and are less prone to over-fitting. So when the amount of data increases, the corresponding network structure should increase. Considering the influence of training time and computing power, according to the size of the datasets B after data augmentation, we choose the median value and take ResNet50 as the pre-trained model in order to get better training effect.\u003c/p\u003e \u003cp\u003eExperimental results of \u003cem\u003eExperiment 1\u0026amp;3\u003c/em\u003e show that the proposed method has achieved excellent performance in the image classification task of DFU, and the augmented data obviously helped to improve the experimental accuracy. After data augmentation, the accuracy of the model on the verification set reached 0.9867 from 0.9167. It showed that the proposed method can accurately classify DFU images.\u003c/p\u003e \u003cp\u003eHowever, images are collected in a standardized environment, and the influence of other complex environments on the accuracy of image classification is not considered.\u003c/p\u003e"},{"header":"V. Conclusion","content":"\u003cp\u003eThe experimental results show that the proposed method has achieved satisfactory performance in the image classification task of DFU, and it is suitable for the general screening of diabetic patients in rural and underdeveloped areas, so as to reduce the cost of their long journey to hospitals or health centers. This study provides new possibilities for the diagnosis and treatment of diabetic patients and is expected to play an important role in clinical practice.\u003c/p\u003e"},{"header":"Declarations","content":"\u003ch2\u003eData Availability Statement\u003c/h2\u003e\n\u003cp\u003eThe raw data supporting the conclusions of this article will be made available by the authors, without undue reservation. Any data or questions regarding the study, please kindly contact Zhen Yu([email protected]) and Cheng Wang (corresponding author,\u0026nbsp;[email protected]).\u003c/p\u003e\n\u003ch2\u003eEthics Statement\u003c/h2\u003e\n\u003cp\u003eThe study was reviewed and approved by the institutional review board of Beijing Chaoyang Hospital affiliated to Capital Medical University (number is 2022-ke-629). All participants have signed informed consent forms.\u003c/p\u003e\n\u003ch2\u003eAuthor Contributions\u003c/h2\u003e\n\u003cp\u003eZhen Yu did the experiments. Zhen Yu, Zhou Long and Cheng Wang analyzed experimental data. Cheng Wang and Hui Zhao designed the research. Zhen Yu and Cheng Wang together as co-author wrote the manuscript, and both are the first authors of the manuscript. All authors contributed to the article and approved the submitted version.\u003c/p\u003e\n\u003ch2\u003eFunding\u003c/h2\u003e\n\u003cp\u003eThis work is partly supported by Shandong Province Science and Technology Small and Medium sized Enterprise Innovation Capability Enhancement Project (2022TSGC2532), 2025 Key Tech-nology Innovation Program of Ningbo(2020Z014) and Capital Health Development Research Special Project (2024-2-7042). The authors are grateful for these supports.\u003c/p\u003e\n\u003ch2\u003eConflict of Interest\u003c/h2\u003e\n\u003cp\u003eThe authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eJiang Y, Wang X, Lei X, et al. A cohort study of diabetic patients and diabetic footulceration patients in China[J]. Wound Repair and Regeneration, 2015,23(2).\u003c/li\u003e\n\u003cli\u003eJiang Qixia, Geng Guangli, Chang Yan, et al. Prevention of foot ulcer in 188 diabetic patients [J]. Chinese Journal of Nursing, 2001(2):85-87.\u003c/li\u003e\n\u003cli\u003eXu Y, Wang L, He J, et al. Prevalence and Control of Diabetes in Chinese Adults.JAMA. 2013;310(9):948\u0026ndash;959. doi:10.1001/jama.2013.168118\u003c/li\u003e\n\u003cli\u003eWang Aihong, Zhao Kun, Li Qiang, et al. Investigation and medical economic analysis of diabetic foot in some provinces and cities of China [J]. chinese journal of endocrinology and metabolism, 2005, 21(6): 496-499.\u003c/li\u003e\n\u003cli\u003eOyibo S O, Jude E B, Tarawneh I, et al. A comparison of two diabetic foot ulcer classification systems: the Wagner and the University of Texas wound classification systems[J]. Diabetes care, 2001, 24(1): 84-88.\u003c/li\u003e\n\u003cli\u003eWang L, Pedersen PC, Agu E, Strong DM, Tulu B. Area determination of diabetic foot ulcer images using a cascaded two-stage SVM-based classification. IEEE Trans Biomed Eng (2016) 64(9):2098\u0026ndash;109. doi: 10.1109/TBME.2016.2632522\u003c/li\u003e\n\u003cli\u003ePatel S, Patel R, Desai D. Diabetic foot ulcer wound tissue detection and classification[C]//2017 international conference on innovations in information, embedded andcommunication systems (ICIIECS). IEEE, 2017: 1-5.\u003c/li\u003e\n\u003cli\u003eYadav MK, Manohar DD, Mukherjee G, Chakraborty C. Segmentation of Chronic Wound Areas by Clustering Techniques Using Selected Color Space[J]. Journal of Medical Imaging and Health Informatics, 2013,3(1): 22-29.\u003c/li\u003e\n\u003cli\u003eGoyal M, Reeves N D, Rajbhandari S, et al. Robust methods for real-time diabeticfoot ulcer detection and localization on mobile devices[J]. IEEE journal of biomedical and health informatics, 2018, 23(4): 1730-1741.\u003c/li\u003e\n\u003cli\u003eChan H P, Samala R K, Hadjiiski L M, et al. Deep learning in medical image analysis[J]. Deep learning in medical image analysis: challenges and applications, 2020:3-21.\u003c/li\u003e\n\u003cli\u003eMin S, Lee B, Yoon S. Deep learning in bioinformatics[J]. Briefings in bioinformatics, 2017, 18(5): 851-869.\u003c/li\u003e\n\u003cli\u003eGoyal M, Reeves ND, Davison AK, Rajbhandari S, Spragg J, Yap MH. Dfunet: Convolutional neural networks for diabetic foot ulcer classification. IEEE Trans Emerg Topic Comput Intell (2018) 4(5):728\u0026ndash;39. doi: 10.1109/TETCI.2018. 2866254\u003c/li\u003e\n\u003cli\u003eAlzubaidi L, Fadhel MA, Oleiwi SR, Al-Shamma O, Zhang J. DFU_QUTNet: diabetic foot ulcer classification using novel deep convolutional neural network. Multimed Tools Appl (2020) 79(21):15655\u0026ndash;77. doi: 10.1007/ s11042-019-07820-w\u003c/li\u003e\n\u003cli\u003eXu Y, Han K, Zhou Y, Wu J, Xie X, Xiang W. Classification of diabetic foot ulcers using class knowledge banks. Front Bioengineer Biotechnol (2021) 9. doi: 10.3389/fbioe.2021.811028\u003c/li\u003e\n\u003cli\u003eCruz-Vega I, Hernandez-Contreras D, Peregrina-Barreto H, RangelMagdaleno JJ, Ramirez-Cortes JM. Deep learning classification for diabetic foot thermograms. Sensors (2020) 20(6):1762. doi: 10.3390/s20061762\u003c/li\u003e\n\u003cli\u003eYap M H, Kendrick C, Reeves N D, et al. Development of diabetic foot ulcer datasets: an overview[J]. Diabetic Foot Ulcers Grand Challenge, 2021: 1-18.\u003c/li\u003e\n\u003cli\u003eHe K, Zhang X, Ren S, et al. Deep residual learning for image recognition[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2016: 770-778.\u003c/li\u003e\n\u003cli\u003eVaswani A, Shazeer N, Parmar N, et al. Attention is all you need[J]. Advances in neural information processing systems, 2017, 30.\u003c/li\u003e\n\u003cli\u003eAhsan M, Naz S, Ahmad R, et al. A deep learning approach for diabetic foot ulcerclassification and recognition[J]. Information, 2023, 14(1): 36.\u003c/li\u003e\n\u003cli\u003eDevries, T., Taylor, G.W.: Improved Regularization of Convolutional Neural Net-works with Cutout. 1708.04552 (2017)\u003c/li\u003e\n\u003cli\u003eZhong, Z., Zheng, L., Kang, G., Li, S., Yang, Y.: Random Erasing Data Augmentation. In: The Thirty-Fourth AAAI Conference on Artifificial Intelligence, AAAI 2020, The Thirty- Second Innovative Applications of Artifificial Intelligence Conference, IAAI 2020, The Tenth AAAI Symposium on Educational Advances in Artifificial Intelligence, EAAI 2020, New York, NY, USA, February 7-12, 2020. pp. 13001\u0026ndash;13008. AAAI Press (2020)\u003c/li\u003e\n\u003cli\u003eTakahashi, R., Matsubara, T., Uehara, K.: RICAP: Random Image Cropping and Patching Data Augmentation for Deep CNNs. In: Zhu, J., Takeuchi, I. (eds.) Proceedings of The 10th Asian Conference on Machine Learning, ACML 2018, Beijing, China, November 14-16, 2018. Proceedings of Machine Learning Research, vol. 95, pp. 786\u0026ndash;798. PMLR (2018)\u003c/li\u003e\n\u003cli\u003eLopes, R.G., Yin, D., Poole, B., Gilmer, J., Cubuk, E.D.: Improving Robustness Without Sacrifificing Accuracy with Patch Gaussian Augmentation. 1906.02611 (2019)\u003c/li\u003e\n\u003cli\u003eZhou, K., Yang, Y., Qiao, Y., Xiang, T.: Domain generalization with mixstyle. CoRR 2104.02008 (2021)\u003c/li\u003e\n\u003cli\u003eBengio Y. Practical recommendations for gradient-based training of deep architectures[M]//Neural networks: Tricks of the trade: Second edition. Berlin, Heidelberg: Springer Berlin Heidelberg, 2012: 437-478.\u003c/li\u003e\n\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":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Diabetes Foot Ulcer, classification method, deep learning","lastPublishedDoi":"10.21203/rs.3.rs-4819913/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-4819913/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground and Objective:\u003c/h2\u003e \u003cp\u003eDiabetes foot ulcer (DFU) is one of the common complications of diabetes patients, which may lead to infection, necrosis and even amputation. Therefore, early diagnosis, classification of severity and related treatment are crucial for the patients. However, traditional DFU classification methods often require experienced doctors to manually classify infections or non-infections, ischemia or non-ischemia, which is time-consuming and low accuracy. The objective of the study is to propose a few-shot DFU image classification method based on deep residual neural network and transfer learning.\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e \u003cp\u003eConsidering the difficulty in obtaining clinical DFU images, it is a few-shot problem. Therefore, the methods include: (1)Data augmentation of the original DFU images by using geometric transformations and random noise; (2)Deep ResNet models selection based on different convolutional layers comparative experiments; (3)DFU classification model training with transfer learning by using the selected pre-trained ResNet model and fine tuning.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003eTo verify the proposed classification method, the experiments were performed with the original and augmented image datasets by separating three classifications:zero grade, mild grade, severe grade. (1) The datasets were augmented from the original 146 to 3000 image datasets and the corresponding DFU classification\u0026rsquo;s average accuracy from 0.9167 to 0.9867; (2)Comparative experiments were conducted with ResNet18, ResNet34, ResNet50, ResNet101, ResNet152 by using 146 image datasets, and the average accuracy/loss is 0.686/0.4649, 0.588/0.5424, 0.549/0.6036, 0.510/1.7144, 0.471/0.5462 respectively; (3) Based on the augmented 3000 image datasets, it was achieved 0.9867 average accuracy with the DFU classification model, which was trained by the pre-trained ResNet50 and hyper-parameters.\u003c/p\u003e\u003ch2\u003eConclusions\u003c/h2\u003e \u003cp\u003eThe experimental results indicated that the proposed few-shot DFU image classification method based on deep residual neural network and transfer learning got very high accuracy, and it can be applied to screening for the diabetes in rural and undeveloped areas.\u003c/p\u003e","manuscriptTitle":"A Few-shot Diabetes Foot Ulcer Image Classification Method Based on Deep ResNet and Transfer Learning","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-09-16 00:41:49","doi":"10.21203/rs.3.rs-4819913/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2024-10-11T08:19:53+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2024-10-08T16:10:15+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2024-08-30T20:21:43+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"81982658921188019656819994487559623196","date":"2024-08-27T16:26:21+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"168334008093260312514322459656432585816","date":"2024-08-27T15:26:45+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2024-08-27T14:20:09+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2024-08-20T09:53:49+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2024-08-08T15:02:36+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2024-08-07T13:18:21+00:00","index":"","fulltext":""},{"type":"submitted","content":"Scientific Reports","date":"2024-07-29T07:29:58+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"1f9c9ad0-221f-4b98-935d-461d7060f86d","owner":[],"postedDate":"September 16th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[{"id":36906710,"name":"Physical sciences/Engineering/Biomedical engineering"},{"id":36906711,"name":"Physical sciences/Mathematics and computing/Computer science"}],"tags":[],"updatedAt":"2024-12-09T16:05:28+00:00","versionOfRecord":{"articleIdentity":"rs-4819913","link":"https://doi.org/10.1038/s41598-024-80691-w","journal":{"identity":"scientific-reports","isVorOnly":false,"title":"Scientific Reports"},"publishedOn":"2024-12-02 15:57:58","publishedOnDateReadable":"December 2nd, 2024"},"versionCreatedAt":"2024-09-16 00:41:49","video":"","vorDoi":"10.1038/s41598-024-80691-w","vorDoiUrl":"https://doi.org/10.1038/s41598-024-80691-w","workflowStages":[]},"version":"v1","identity":"rs-4819913","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-4819913","identity":"rs-4819913","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}