MUMBAI, India, Sept. 22 -- Intellectual Property India has published a patent application (202641110464 A) filed by K Thilak Sekhar on September 15, 2026, for Toothai.
Inventors include K Thilak Sekhar; Prof. D. Jagadeesan; and Dr. Lokesh Ravi.
The application for the patent was published on September 18, 2026, under issue no. 38/2026.
Abstract: The present invention provides a computer implemented system and method for automated, corruption aware detection, severity classification and tooth level clinical mapping of periapical lesions from panoramic dental radiographs. The system validates and auto corrects out-of-bound, invalidly ordered and augmentation corrupted bounding box annotations prior to model training, applies contrast limited adaptive histogram equalisation, and processes the radiograph as a plurality of overlapping high resolution image tiles to preserve small lesion detail. A lesion detection model, trained using a class imbalance aware objective combining focal loss and severity grade oversampling, detects periapical lesions and their severity grade within the tiles. During inference, a sliding window mechanism applies the model across the tiles and a plurality of test time augmented variants, and the resulting candidate detections are fused using soft non-maximum suppression with Gaussian score decay. A segmentation module generates separate tooth body and lesion-masks, the tooth body mask being anatomically extended according to jaw position and crown orientation. A mapping module assigns an FDI tooth number to each detected lesion, including a safeguard for a modulo based numbering edge case, and a report generation module combines the lesion, its severity grade, its FDI tooth number and a reliability indicator into a structured clinical output.With overall precision of 85%, sensitivity of 70%, and mAP@50 of 80% on the held-out test set, YOLOv8l showed the most consistent convergence and the best detection performance.
Disclaimer: Curated by HT Syndication.
