MUMBAI, India, Aug. 17 -- Intellectual Property India has published a patent application (202641095252 A) filed by Adhiparasakthi Engineering College, Melmaruvathur on August 06, 2026, for Ai-Powered Hybrid Vision Transformer–lstm Model For Sperm Morphology Analysis Using Image And Numeric Data With Web-Based Prediction.
Inventors include Dr. C. Dhaya; R. Swathy; Dr. R. Srivel; Ms. V. Thamarai Selvi; Ms. A. Jayanthi; and Mrs. Devika M.
The application for the patent was published on August 14, 2026, under issue no. 33/2026.
Abstract: Sperm morphology analysis is a critical determinant of male fertility, yet traditional manual microscopic evaluation is time-consuming, subjective, and inconsistent, while existing computer-aided sperm analysis (CASA) systems remain limited by poor noise handling, single-modality processing, and unoptimized model parameters. This invention proposes an Optimized Spatio-Temporal Graph Attention Framework for multi-modal sperm morphology assessment that integrates image and numeric clinical data through parallel processing pathways. Sperm images are denoised using Savitzky–Golay polynomial filtering, segmented via Gray-Level Co-occurrence Matrix entropy-based texture region growing, and characterized using Scale-Invariant Feature Transform (SIFT) descriptors, while numeric morphometric and clinical parameters undergo cleaning, encoding, correlation-based visualization, and normalization. Both feature streams are fused as graph-structured inputs to a Spatio- Temporal Graph Attention Network (ST-GAT) comprising graph-convolutional layers, positional embeddings, multi-head self-attention transformer encoders, and an LSTM temporal aggregation layer, with hyperparameters automatically tuned via an Ant Lion Optimizer. The system classifies samples into Normal Sperm, Abnormal Sperm, and Non-Sperm categories, achieving up to 100% accuracy on the numeric pathway and 89% accuracy on the image pathway, and is deployed through a Flask- based web application offering real-time image-upload and numeric-parameter prediction interfaces with derived health-percentage indicators. The proposed framework delivers an objective, robust, and readily deployable solution for automated sperm morphology assessment, directly supporting fertility diagnostics, assisted reproductive treatment planning, and reproductive health research
Disclaimer: Curated by HT Syndication.
