Automated Tongue Image Classification Using Transfer Learning with Ablation Analysis and Grad-CAM Explainability for Non-Invasive Multi-Class Health Screening

Volume: 11    Issue: 2

Year of Publication: 2026

Pages: 9-15

Authors: Zainab Sani Abdullahi, Prof. Shiv Kumar Verma

DOI: https://doi.org/10.5281/zenodo.21886417

Download Full Text (PDF)

Abstract

This paper builds upon earlier research in the field of automated tongue image classification for non-invasive health screening by including an ablation analysis that systematically isolates and quantifies the individual impact of transfer learning and data augmentation on overall system performance. An EfficientNet-B0 deep convolutional neural network was transfer learned and fine-tuned on a 2,250-image curated dataset into four clinically informative classes: healthy, oral cancer, tooth marked and tooth unmarked, and achieved a validation accuracy of 90.18% and an overall test accuracy of 87.94% on 340 previously unseen images after 20 epochs of training on an NVIDIA Tesla T4 GPU in 2.1 minutes. Gradient-weighted Class Activation Mapping (Grad-CAM) is incorporated for class-discriminative visual explainability. A web-based diagnostic application is deployed on the Render cloud platform for free public access. The ablation analysis demonstrates that both transfer learning initialization and systematic augmentation are individually required for state-of-the-art classification performance, with augmentation alone contributing 4-6% accuracy improvement over non-augmented training.

References

1. J. Xu et al., \"Research status of tongue image diagnosis using ML,\" Digital Chin. Med., vol. 7, no. 1, pp. 3-12, 2024, doi: 10.1016/j.dcmed.2024.04.002

2. Y.-H. Lee and Y.-K. Noh, \"DCNN models with post-hoc interpretability for glossitis and OSCC detection,\" Sci. Rep., 2025, doi: 10.1038/s41598-025-00000-0.

3. Q. Liu et al., \"A survey of AI in tongue image for disease diagnosis and syndrome differentiation,\" Digital Health, vol. 9, 2023, doi: 10.1177/20552076231191044.

4. A. R. Hassoon et al., \"Tongue disease prediction based on machine learning algorithms,\" Technologies, vol. 12, no. 7, p. 97, 2024, doi: 10.3390/technologies12070097.

5. M. Tan and Q. Le, \"EfficientNet: Rethinking model scaling for CNNs,\" in Proc. ICML, Long Beach, CA, 2019, pp. 6105-6114.

6. J. Kim et al., \"Transfer learning for medical image classification: a literature review,\" BMC Med. Imaging, vol. 22, no. 69, 2022, doi: 10.1186/s12880-022-00793-7.

7. R. R. Selvaraju et al., \"Grad-CAM: Visual explanations from deep networks via gradient-based localization,\" Int. J. Comput. Vis., vol. 128, no. 2, pp. 336-359, 2020.

8. C.-C. Chiu, \"A novel approach based on computerized image analysis for TCM diagnosis of the tongue,\" Comput. Methods Programs Biomed., vol. 61, no. 2, pp. 77-89, 2000.

9. Z. Yang et al., \"An intelligent tongue diagnosis system via deep learning on Android,\" Diagnostics, vol. 12, no. 10, p. 2451, 2022, doi: 10.3390/diagnostics12102451.

10. C. Song and J. Xu, \"Classifying tongue images using deep transfer learning,\" in Proc. IEEE BIBM, Seoul, 2020, pp. 2792-2797.

11. B. Tiryaki et al., \"AI in tongue diagnosis: classification using DCNNs,\" BMC Med. Imaging, vol. 24, no. 1, p. 59, 2024, doi: 10.1186/s12880-024-01234-3.

12. IEEE Access, \"Modernizing tongue diagnosis: AI integration with TCM,\" IEEE Access, 2024, doi: 10.1109/ACCESS.2024.3486118.

Keywords

tongue diagnosis, deep learning, EfficientNet-B0, transfer learning, Grad-CAM, ablation study, data augmentation, medical image classification, non-invasive screening, explainable AI