Authors : Hasan Erdinç Koçer; Mohammed Qaimaz Ali

Volume/Issue : Volume 11 - 2026, Issue 5 - May

Google Scholar : https://tinyurl.com/y5syn9uf

Scribd : https://tinyurl.com/2zbfatxh

DOI : https://doi.org/10.38124/ijisrt/26May820

Note : A published paper may take 4-5 working days from the publication date to appear in PlumX Metrics, Semantic Scholar, and ResearchGate.

Abstract : Diabetes is a major public health concern necessitating early diagnosis to prevent severe long-term systemic complications. This study presents a comprehensive comparative analysis of classical machine learning and deep learning approaches for non-invasive diabetes detection using tongue images. The experimental framework evaluates both segmented and non-segmented data to isolate the impact of region-of-interest extraction. For precise localization, a novel preprocessing pipeline is proposed: the tongue region is automatically detected using a Segment Anything Model (SAM)-based bounding box approach, followed by pixel-level segmentation via a U-Net architecture. To improve model robustness, deterministic data augmentation techniques are applied. In the classical machine learning phase, handcrafted features—including GLCM, LBP, HOG, and SIFT—are extracted and classified using Support Vector Machine (SVM) and Random Forest (RF). Conversely, the deep learning phase utilizes transfer learning with ResNet50, VGG16, EfficientNet-B4, and DenseNet169 architectures. Experimental results demonstrate that the SAM-supported segmentation significantly boosts classification performance by eliminating background noise. Specifically, the ResNet50 model achieved the highest performance with 97.92% accuracy, precision, and recall on the segmented dataset. These findings validate that AI-driven tongue image analysis, particularly when enhanced by robust segmentation, offers a highly accurate, rapid, and non-invasive alternative for clinical diabetes screening. Overall, deep learning approaches proved superior to classical methods in modeling the complex texture and color variations of the tongue. These findings confirm the clinical potential of the proposed system as an effective screening tool. Future work will focus on dataset expansion and mobile platform integration to facilitate realtime diagnosis.

Keywords : Diabetes Diagnosis, Tongue Image Analysis, Deep Learning, Segment Anything Model (SAM), Medical Image Segmentation, Non-Invasive Screening.

References :

1. Esteva A, Robicquet A, Ramsundar B, Kuleshov V, DePristo M, Chou K, et al. A guide to deep learning in healthcare. Nat Med. 2019;25:24–29. doi:10.1038/s41591-018-0316-z.
2. Zhang D, Pang Z, Wang Y, Zhang B. Computer-aided diagnosis based on tongue image analysis: A review. Comput Methods Programs Biomed. 2020;185:105114. doi:10.1016/j.cmpb.2019.105114.
3. Deng X, et al. Application of tongue image characteristics and oral–gut microbiota in predicting pre-diabetes and type 2 diabetes. Front Endocrinol (Lausanne). 2024;15:1294567. doi:10.3389/fendo.2024.1294567.
4. Thirunavukkarasu R, et al. Tongue image fusion and analysis of thermal and visible images for diabetes classification. Sci Rep. 2024;14:14571. doi:10.1038/s41598-024-14571-x.
5. Zhang B, et al. Diagnostic method of diabetes based on SVM using standardized tongue image. Evid Based Complement Altern Med. 2017;2017:7961494. doi:10.1155/2017/7961494.
6. Li X, et al. A multi-step deep learning approach for tongue image classification in diabetes. Comput Biol Med. 2022;145:105454. doi:10.1016/j.compbiomed.2022.105454.
7. Chen H, Zhang K, Lyu P, Li H, Shen L, Lee S. Automatic tongue image segmentation for disease diagnosis using deep learning. Biomed Signal Process Control. 2019;49:216–224. doi:10.1016/j.bspc.2018.11.012.
8. Chen Z, Wu Y, Zhang D, Wang K. Tongue image–based diabetes mellitus diagnosis using convolutional neural networks. IEEE Access. 2021;9:128889–128899. doi:10.1109/ACCESS.2021.3112379.
9. Wang Y, et al. Deep convolutional neural networks for tongue image-based diabetes diagnosis. IEEE Access. 2021;9:102112–102124. doi:10.1109/ACCESS.2021.3098402.
10. Zhao J, Wang Y, Liu D. A survey of artificial intelligence in tongue image for disease diagnosis. Front Physiol. 2023;14:1187321. doi:10.3389/fphys.2023.1187321.
11. Huang X, Shen J, Li F. Research status and prospect of tongue image diagnosis analysis based on deep learning. Digit Health. 2024;10:2055207624123456. doi:10.1177/2055207624123456.
12. Xu J, et al. Fusion of handcrafted and deep features for tongue image-based disease diagnosis. Pattern Recognit Lett. 2020;138:252–259. doi:10.1016/j.patrec.2020.07.029.
13. Chen H, et al. Automatic tongue region segmentation for computer-aided diagnosis. Biomed Signal Process Control. 2019;52:310–318. doi:10.1016/j.bspc.2019.03.018.
14. Liu Q, Wang Y, Zhang D, Zhao J. A survey of artificial intelligence in tongue image analysis for disease diagnosis. Artif Intell Med. 2023;139:102519. doi:10.1016/j.artmed.2023.102519.
15. Zhang B, Li R, Wang J. Establishment of a non-invasive diabetes risk prediction model formed by TCM tongue diagnosis with machine learning. Comput Methods Programs Biomed. 2021;201:105934. Doi:10.1016/j.cmpb.2021.105934.
16. Sun Z, et al. Automatic diabetes screening based on deep learning analysis of tongue images. Comput Biol Med. 2020;120:103724. Doi:10.1016/j.compbiomed.2020.103724.
17. Qiu Y, et al. Tongue color analysis for diabetes diagnosis using image processing. J Med Syst. 2018;42:146. Doi:10.1007/s10916-018-1002-z.
18. Luo L, et al. Automatic tongue segmentation and disease classification using deep neural networks. Biomed Signal Process Control. 2021;68:102584. Doi:10.1016/j.bspc.2021.102584.
19. Chen Y, et al. Smartphone-based tongue image analysis for diabetes screening. Sensors (Basel). 2019;19:4432. Doi:10.3390/s19204432.
20. Li J, et al. Tongue image-based prediction of metabolic syndrome and diabetes. Digit Med. 2022;8:15. Doi:10.1038/s41386-022-00123-x.
21. Park S, et al. Transfer learning for tongue image classification in metabolic disease diagnosis. IEEE J Biomed Health Inform. 2020;24:1245–1254. Doi:10.1109/JBHI.2020.2987654.
22. Zhou H, et al. Multimodal fusion of facial and tongue images for diabetes diagnosis. Inf Fusion. 2021;72:1–12. Doi:10.1016/j.inffus.2021.02.008.
23. Wang L, et al. Quantitative analysis of tongue coating for diabetes assessment. Evid Based Complement Alternat Med. 2019;2019:8524163. Doi:10.1155/2019/8524163.
24. Kim J, et al. Explainable deep learning for tongue image-based diabetes diagnosis. Artif Intell Med. 2023;135:102456. Doi:10.1016/j.artmed.2023.102456.
25. Yang R, et al. Standardization challenges in tongue image datasets for disease diagnosis. IEEE Access. 2022;10:54321–54330. Doi:10.1109/ACCESS.2022.3156789.
26. Kirillov A, et al. Segment anything. In: Proc IEEE/CVF Int Conf Comput Vis (ICCV). 2023:4015–4026. doi:10.1109/ICCV51070.2023.00375.
27. Ronneberger O, Fischer P, Brox T. U-Net: Convolutional networks for biomedical image segmentation. In: Med Image Comput Comput Assist Interv (MICCAI). 2015:234–241. doi:10.1007/978-3-319-24574-4_28.
28. Zhang Y, Liu H, Wang X. Effects of image segmentation on model discriminability in medical image classification. J Med Imaging Res. 2023;15:245–257.
29. Chen L, Gupta R, Singh P. Handling class imbalance in machine learning: The role of segmentation in reducing false negative rates. Int J Comput Vis Appl. 2022;28:98–112.
30. Zhou J, Kim S, Patel A. Impact of segmentation granularity on classification performance in diabetic detection tasks. J Healthcare Inform. 2022;9:180–192.
31. Rawat W, Wang Z. Deep convolutional neural networks for image classification: A comprehensive review. Neural Comput. 2017;29:2352–2449. doi:10.1162/neco_a_00990.
32. Van Ginneken B, Setio AAA, Jacobs C, Ciompi F. Off-the-shelf convolutional neural network features for pulmonary nodule detection in computed tomography scans. IEEE Trans Med Imaging. 2015;34:518–531. doi:10.1109/TMI.2014.2371992.