MUMBAI, India, Oct. 26 -- Intellectual Property India has published a patent application (202641123920 A) filed by Vellore Institute Of Technology on October 16, 2026, for Asymmetric Temporal-To-Static Knowledge Distillation For Lightweight Thyroid Nodule Malignancy Prediction From Ultrasound.
Inventors include Bhawana Tyagi; Ayaan Sehgal; Monami Som Saha; and Vedant Vijay Zade.
The application for the patent was published on October 23, 2026, under issue no. 43/2026.
Abstract: The present invention provides an Asymmetric Temporal-to-Static Knowledge Distillation (AT2S-KD) system and method for lightweight thyroid nodule malignancy prediction using ultrasound cine-loops and static ultrasound images. The system comprises a spatio-temporal teacher network configured to learn temporal diagnostic embeddings from thyroid ultrasound cine-loops and a lightweight static-frame student network configured to receive individual ultrasound frames. A Temporal Embedding Alignment Loss (TEAL) framework transfers temporal knowledge by jointly optimizing supervised classification, normalized embedding alignment, contrastive anti-collapse regularization, and ACR TI-RADS multi-task learning. A Pseudo-Temporal Simulation (PTS) mechanism enables distillation from unpaired static images by generating pseudo-temporal sequences and obtaining temporal embedding targets from a frozen teacher network. The student network further provides malignancy prediction, TI-RADS classification, nodule segmentation, and carcinoma subtype classification. The system provides calibrated prediction and deployment through an ONNX-compatible model configured for 320×320 input resolution and CPU and/or GPU execution.
Disclaimer: Curated by HT Syndication.
