Authors: Anjali Kadam, Aishwarya Bhosale, Vaishnavi Jadhav, Swara Chavan, Dnyaneshwari Mohotkar
Abstract: Tongue diagnosis has traditionally been a non-invasive method for detecting gastrointestinal (GI) disorders, widely practiced in Eastern medicine. This research explores the use of a fine-tuned ResNet50 model for tongue image classification to aid in the diagnosis of gastrointestinal (GI) disorders. The model was trained on labeled images focused on three conditions: fissure, constipation, and hyperacidity. The dataset was manually collected from patients with assistance from an Ayurvedic practitioner, including hospital visits and shared tongue images. Preprocessing and augmentation techniques were applied to enhance generalization. The model achieved 80–97% accuracy on known images but dropped to 50–60% on unseen data, highlighting the need for a larger dataset. This project is intended as a foundation for future research, with the expectation that
the accuracy and number of diagnosable conditions will improve as the dataset expands.
DOI: http://doi.org/
