Transfer learning for classyfying otoscopic medical images in the diagnosis of ent diseases

Relevance. According to the Ministry of Health of the Russian Federation, more than 6% of the country's population consults an otorhinolaryngologist annually due to ear diseases. Given the high workload of specialists, the accelerated implementation of artificial intelligence (AI) methods to assist in diagnosis is necessary. Objective. To develop a prototype AI-based medical system (AI MS) for diagnosing ear diseases from otoscopic images (OIs), achieving a diagnostic accuracy of at least 95% on the test set through transfer learning techniques applied to pre-trained deep neural networks (NNs).

Materials and methods. A labeled otoscopic image dataset representing five balanced diagnostic classes was selected and prepared from Kaggle.com. To effectively implement transfer learning of neural network models and prevent overfitting, a data preprocessing and augmentation pipeline was established. A custom web interface for NNM integration was developed, and the Grad-CAM method was applied to visualize diagnostic regions within the images.

Results. The CNN models selected for transfer learning were ResNet18, ResNet50, EfficientNet-B0, and EfficientNet-B2. Based on experiments using "soft" augmentation of the training set, the diagnostic accuracy for five ear disease categories—listed in the order mentioned—was 95.3%, 96.1%, 97.8%, and 99.7%, respectively. The ResNet50 model with soft augmentation of the training set for 3 classes showed an overall test data accuracy of 97%. Heatmaps visualizing the diagnostic results using the Grad-CAM method are presented.

Conclusion. Transfer learning, involving the adaptation of selected pre-trained architectures (ResNet18, ResNet50, EfficientNet-B0, and EfficientNet-B2) to the task of multiclass classification of ear diseases using balanced classes, enabled the achievement of the target accuracy for automated diagnosis. Conditions have been established for collaborative efforts with the Mogilev Regional Clinical Hospital (MRCH) to develop an AI-based medical system for diagnosing ear diseases.

1. Ukaz Prezidenta RF №124 ot 15.02.2024 g. «O vnesenii izmenenij v Ukaz Prezidenta Rossijskoj Federacii ot 10 oktyabrya 2019 g. №490 «O razvitii iskusstvennogo intellekta v Rossijskoj Federacii» i v Nacional'nuyu strategiyu, utverzhdennuyu etim Ukazom». Available at: http://publication.pravo.gov.ru/document/0001202402150063. Accessed 04.12.2025. (In Russ.)

2. National Legal Internet Portal of the Republic of Belarus. Artificial Intelligence in Belarus: what technologies are being developed and how they will be regulated. 09.06.2025. Available at: https://pravo.by/novosti/obshchestvenno-politicheskie-i-v-oblastiprava/2025/june/88987/ Accessed 04.12.2025. (In Russ.)

3. Like PIROGOV.AI uses AI to solve the problem of waiting lists for an otolaryngologist. PIROGOV.AI is a system for establishing expert diagnosis of ENT diseases and supporting medical decision-making on endoscopic photo and video images. RBC Trends. Available at: https://trends.rbc.ru/trends/industry/5f97deb09a7947213445eb5f?ysclid=ml6fj5pld112736064. Accessed 04.02.2026. (In Russ.)

4. Khublaryan AG, Kryukov AI, Kunelskaya NL, et al. Application of artificial intelligence algorithms for diagnosing the pathology of ear diseases. Digital Diagnostics. 2024; 5(1S): 102-105. (In Russ.) doi: 10.17816/DD627081.

5. Kryukov AI, Garov EV, Sudarev PA, et al. Performance evaluation results of an artificial neural network developed for the purpose of classifying otoendoscopic images. Russian Bulletin of Otorhinolaryngology. 2025; 90(3): 9‑12. (In Russ.) doi: 10.17116/otorino2025900319.

6. Zeng X, Jiang Z, Luo W, et al. Efficient and accurate identification of ear diseases using an ensemble deep learning model. Scientific Reports. 2021; 11: 10839. (In Russ.) doi: 10.1038/s41598-021-90345-w.

7. Cha D, Pae Ch, Seong S, et al. Automated diagnosis of ear disease using ensemble deep learning with a big otoendoscopy image database. eBioMedicine. 2019; 45: 606-614. (In Russ.)

8. GOST R 71671—2024. Medical decision support systems using artificial intelligence. Moscow: Rosstandart, 2024. Date of introduction – 2025- 01-01. Available at: https://e-stomatology.ru/info/399/Accessed 05.02.2026. (In Russ.)

9. Khryashchev VV. A medical decision support system using an artificial intelligence module for endoscopic gastric examinations // Medical Technology. 2023; 6: 44-47. (In Russ.)

10. Quality control. The accuracy of diagnosis of diseases of the ear, throat, nose. Pirogov.AI Available at: https://pirogov.ai/kontrol-kachestva. Accessed 05.02.2026. (In Russ.)

11. Chibisova MA, Ermolaeva LA, et al. The use of artificial neural networks in the recognition of X-ray image data in outpatient dentistry. REJR. 2025; 15(1): 41-61. (In Russ.) doi: 10.21569/2222-7415-2025-15-1-41-61.

12. How artificial intelligence will change medicine by 2050. July 30, 2025. Available at: https://science.mail.ru/articles/5921-ii-v-medicine-2050/?ysclid=ml7wixgcaq947323240. Accessed 05.02.2026. (In Russ.)

13. Trufanov GE, Efimtsev AY. Artificial intelligence technologies in MR-neuroimaging. The radiologist's View. Russian Journal of Personalized Medicine. 2023; 3(1): 6-7. (In Russ.) doi: 10.18705/2782-3806-2023-3-1-6-17.

14. Bobrovskaya TM, Vasilev YuA, Nikitin NYu, Arzamasov КM. Approaches to building radiology datasets. Medical Doctor and Information Technologies. 2023; 4: 14-23. (In Russ.) doi: 10.25881/18110193_2023_4_14.

15. Kalinin DI, Karyakin AA, Zakharchenko UR, Morozova DA, et al. Creation of an intelligent eardrum image analysis system for differential diagnosis of ear diseases. Bulletin of the Northern State Medical University. 2023; 2: 35-40. (In Russ.)

16. Kogan EL. Ear diseases (Ear diseases): types, symptoms, diagnosis. Polyclinic. Available at: https://polyclin.ru/articles/zabolevaniya-ushei/?ysclid=ml99oqb8bg20379837255. Accessed 05.02.2026. (In Russ.)

17. Methods of diagnosis of hearing diseases. The first clinic in Izmailovo. Available at: https://dr-tb.ru/otolaringologija/metody-diagnostiki-zabolevanij-sluha?ysclid=miswuxm s9w554790961. Accessed 05.12.2025. (In Russ.)

18. What is an Otoscopy? When is it used? Medical center "Ent Plus". Available at: https://lor-plus.ru/otoskopiya?ysclid=misq8e53qn209502860. Accessed 04.12.2025. (In Russ.)

19. Zotina NI. What is otoscopy? Megapteka.ru. Published on 09.10.2025. Available at: https://megapteka.ru/specials/chto-takoe-otoskopiya-6479. Accessed 04.12.2025. (In Russ.)

20. Kristin Hayes. Myringosclerosis, tympanosclerosis and ear health. Health From A to Z Updated on December 10, 2025. Available at: https/www.verywellhealth.com/myringosclerosis-and-tympanosclerosis-1191943. Accessed 04.12.2025. (In Russ.)

21. Mikhailov AA. Tympanosclerosis – symptoms and treatment. The publication date is September 18, 2025. Available at: https://probolezny.ru/timpanoskleroz. Accessed 04.12.2025. (In Russ.)

22. Transfer training: what is it, how does it work, and what is it for? Skillbox Media. February 14, 2025. Available at: https://skillbox.ru/media/code/transfernoe-obuchenie-chto-eto-kak-rabotaet-i-dlya-chego-nuzhno/?ysclid=mirmdg8rw1107140364. Accessed 05.12.2025. (In Russ.)

23. Chernitskaya О. A complete guide to data augmentation for computer vision. Medium. June 1, 2021. Available at: https://medium.com/data-science/complete-guide-to-data-augmentation-for-computer-vision-1abe4063ad07 Accessed 04.02.2026. (In Russ.)

24. Abhishek S. Fine-tuning of the pre-trained ResNet-18 model for image classification on a custom dataset using PyTorch. Medium. September 10, 2024. Available at: https://medium.com/@imabhi1216/fine-tuning-a-pre-trained-resnet-18-model-for-image-classification-on-custom-datasetwith-pytorch-02df12e83c2c. Accessed 08.12.2025. (In Russ.)

25. Abirami V. Accuracy vs. precision vs. recall in machine learning. Ultralytics. August 20, 2025. Available at: https://www.ultralytics.com/ru/blog/accuracy-precision-recall. Accessed 04.02.2026. (In Russ.)

26. A set of otoscopic image data. Kaggle. Available at: https://www.kaggle.com/datasets/ucimachinelearning/otoscopic-image-dataset. Accessed 08.12.2025. (In Russ.)

27. Berezovsky V, Vygovskaya N. An overview of neural networks for medical image recognition. E3S Web of Conferences. 2023; 460: 04028. (In Russ.) doi: 10.1051/e3sconf/202346004028.

28. Neural networks for medical image processing. Everest Solution. 18.01.2025. Available at: https://everest-solution.com/articles/nejrosetidlya-obrabotki-mediczinskih-izobrazhenij/?ysclid=mlb1it1w9y117429784. Accessed 04.02.2026. (In Russ.)

29. Artificial intelligence in medical diagnostics: accuracy and speed. Skypro. Available at: https://sky.pro/wiki/profession/analiz-medicinskih-izobrazhenij-s-pomoshyu-ii. Accessed 04.02.2026. (In Russ.)

30. Krishnakumar R, Tamilselvi K. AI-Powered Otoscopic Image Classification for Ear Disease Detection. Int. J. Sci. Res. Comput. Sci. Eng. Inf. Technol. 2025; 11(4): 1-10. (In Russ.) doi: 10.32628/CSEIT25113394.

31. Convolutional Neural Networks (CNN): A complete guide to how they work, learn, and apply. Kvantech blog. Available at: https://kvan.tech/blog/svertochnye-neyronnye-seti-cnn-polnoerukovodstvo-po-printsipam-raboty-obucheniyu-i-primeneniyu. Accessed 05.12.2025. (In Russ.)

32. A. J. V, J. S, J. A. Kovilpillai J, A. G, A. K. Otoscopy Image Classification Using Embedded AI // IEEE International Conference on Information Technology, Electronics and Intelligent Communication Systems. (ICITEICS). 2024. Available at: https://ieeexplore.ieee.org/abstract/document/10625385. Accessed 04.02.2026. (In Russ.)

33. Afifi HM, Mohammed KK, Hassanen AE. Insight into Automatic Image Diagnosis of Ear Conditions Based on Optimized Deep Learning Approach. Ann Biomed Eng. 2024; 52(4): 865-876. (In Russ.) doi: 10.1007/s10439-023-03422-8

34. Guo K, Xie L, Zhou L. Deep learning for otitis media classification using otoscopic image. Medicine (Baltimore). November 28, 2025. Available at: https://pmc.ncbi.nlm.nih.gov/articles/PMC12662458. Accessed 04.02.2026. (In Russ.) doi: 10.1097/MD.0000000000046218.

35. Tsutsumi K, Goshtasby K, Risboud A, et al. A Web-Based Deep Learning Model for Automated Diagnosis of Otoscopic Images. Otol Neurotol. Available at: https://pmc.ncbi.nlm.nih.gov/articles/PMC8448915. Accessed 05.02.2026. (In Russ.)

36. Ashish J. Understanding EfficientNet Architecture – The Perfect Balance of Depth, Width & Resolution. Medium. October 16, 2025. Available at: https://medium.com/@mechatronics420/understanding-efficientnet-architecture-the-perfect-balance-ofdepth- width-resolution-809b35b5346c. Accessed 05.02.2026. (In Russ.)

37. Efficientnet Architecture. GeeksforGeeks. July 23, 2025. Available at: https://www.geeksforgeeks.org/computervision/efficientnet-architecture. Accessed 05.02.2026. (In Russ.)

38. Choi Yi, Chae J. et al. Automated multi-class classification for prediction of tympanic membrane changes with deep learning models. PLoS One. 2022; 17(10). (In Russ.) doi: 10.1371/journal.pone.0275846.

39. Akyol K. Comprehensive comparison of modified deep convolutional neural networks for automated detection of external and middle ear conditions. 2024; 36: 5529-5544. (In Russ.) doi: 10.1007/s00521-023-09365-4.

40. Pranshu A. Understanding Vision Transformers: A Detailed Guide. Medium. June 18, 2024. Available at: https://medium.com/@pranshu1902/understanding-vision-transformers-a-detailed-guide-c836f7df9ebc. Accessed 04.02.2026. (In Russ.)

41. Swin Transformer. GeeksforGeeks. 13 августа 2025. Доступно по: https://www.geeksforgeeks.org/computer-vision/swin-transformer. Дата обращения: 04.02.2026. [Swin Transformer. GeeksforGeeks. August 13. 2025. Available at: https://www.geeksforgeeks.org/computer-vision/swin-transformer. Accessed 04.02.2026. (In Russ.)]

42. Ebrahimzadeh M, Asadi H. How to Use Transformers for Transfer Learning? May 31, 2023. Available at: http://dx.doi.org/10.2139/ssrn.4461376. Accessed 04.02.2026. (In Russ.)

43. Rehman U, et al. Efficient and Interpretable Otoscopic Image Classification via Distilled CNN With Adaptive Channel Attention. IEEE. 2025; 13: 151082-151096. (In Russ.)

44. Knowledge Distillation. What is Knowledge Distillation? Ultralytics. Available at: https://www.ultralytics.com/ru/glossary/knowledge-distillation. Accessed 04.02.2026. (In Russ.)

45. Dong Kyung K. Grad-CAM: A gradient approach to explainability in deep learning. Medium. February 17, 2025. Available at: https://medium.com/@kdk199604/grad-cam-a-gradient-based-approach-toexplainability-in-deep-learning-871b3ab8a6ce. Accessed 08.12.2025. (In Russ.)

46. Ankit S. Detailed understanding and implementation guide ResNet. Available at: https://cv-tricks.com/keras/understand-implement-resnets. Accessed 08.12.2025. (In Russ.)

47. Navanit M. An introduction to Resnet. January 31, 2024. Available at: https://www.scaler.com/topics/residual-networks-resnetdeep-learning. Accessed 08.12.2025. (In Russ.)

48. Abhik S. Skipping connections. Skip Connections: the ResNet revolution. April 1, 2024. Available at: https://www.abhik.ai/concepts/deep-learning/skip-connections. Accessed 04.02.2026. (In Russ.)

49. ImageNet. Data set, usage, and examples. Available at: https://www.ultralytics.com/ru/glossary/imagenet. Accessed 05.12.2025. (In Russ.)

50. Shouke Wei. Split Dataset for Model Development with Python Scikit-learn. Medium. Nov 17, 2022. Available at: https://medium.com/@shouke.wei/split-dataset-for-model-development-with-python-scikitlearn-7fa83e115981. Accessed 06.02.2026. (In Russ.)

51. Metrics for evaluating the performance of machine learning models. Gimal-Ai. August 18, 2025. Available at: https://gimalai.ru/blog/metriki-otsenki-proizvoditelnosti-modeley-mashinnogo-obucheniya/?ysclid=miwtzbej oi726340652. Accessed 08.12.2025. (In Russ.)

52. Epoch, butch, iteration – what are the differences? Basic course. Neurohiv. December 27, 2018. Available at: https://neurohive.io/ru/osnovy-datascience/jepoha-razmer-batcha-iteracija/?ysclid=mj2fkffkp6514066716. Accessed 08.12.2025. (In Russ.)

53. Google Colaboratory. Available at: https://colab.google. Accessed 08.12.2025. (In Russ.)

54. Computer Vision Model EarAbnormalities. Universe. Available at: https://universe.roboflow.com/cso-nhwes/earabnormalities-togvt. Accessed 11.02.2026. (In Russ.)

55. AbiramiVina. The ultimate guide to data augmentation in 2025. Ultralytics. February 14, 2025. Available at: https://www.ultralytics.com/ru/blog/the-ultimateguide-to-data-augmentation-in-2025. Accessed 11.02.2026. (In Russ.)

56. Yang L, et al. A soft addition for image classification. Arxiv. January 23, 2024. Available at: https://arxiv.org/html/2211.04625v2. Accessed 09.02.2026. (In Russ.)

57. Fagerland MW, Lydersen S, Laake P. Recommended confidence intervals for two independent binomial proportions // Stat Methods Med Res. 2015; 24(2): 224-54. (In Russ.) doi: 10.1177/0962280211415469.

58. The Wilson Confidence interval for a Proportion. Econometrics.blog. February 5, 2022. Available at: https://www.econometrics.blog/post/the-wilson-confidence-intervalfor-a-proportion.Accessed 18.03.2026. (In Russ.)

59. Vygovskaya NV. LOR_Model_Web: a computer program for the diagnosis of otitis media of the middle ear using otoscopic images using artificial intelligence. Registered in the Register of Computer Programs on 11/14/2025, №1941. The deposit period is until: 11/02/2027. Available at: https://search.ncip.by/depon/index.php?pref=1&lng=ru&page=3&target=2391. Accessed 08.12.2025. (In Russ.)

60. FastAPI. Documentation FastAPI. Available at: https://rhkina.gitlab.io/fastapi. Accessed 08.12.2025. (In Russ.)]

61. Louis M. Explainability of AI using Grad-CAM: visualization of neural network solutions. January 8, 2025. Available at: https://edgeimpulse.com/blog/ai-explainability-with-grad-cam-visualizing-neural-network-decisions. Accessed 08.12.2025. (In Russ.)

62. Shchukina NA. Neural network models in the problem of medical images classification. Modeling, Optimization and Information Technology. 2021; 9(4). (In Russ.) doi: 10.26102/2310-6018/2021.34.3.022.

63. Kozar RV, Konoiko NS, Navrotsky AA. Data clustering methods for recognition of endoscopic image in the problems of computer medical diagnosis. BGUIR. 2023. 21(1): 94-97. (In Russ.) doi: 10.35596/1729-7648-2023-21-1-94-97.