MUMBAI, India, Aug. 17 -- Intellectual Property India has published a patent application (202641096828 A) filed by Sanjiv Kumar; Dr. Ankur Dumka; and Dr. Sushil Kumar on August 11, 2026, for A System And Method For Domain-Constrained Adaptive Hyperparameter Optimization In Deep Learning-Based Brain Tumor Segmentation.
Inventors include Sanjiv Kumar; Dr. Ankur Dumka; and Dr. Sushil Kumar.
The application for the patent was published on August 14, 2026, under issue no. 33/2026.
Abstract: The present invention discloses a system and method for domain-constrained adaptive hyperparameter optimization in deep learning-based brain tumor segmentation. The system receives brain MRI images and corresponding segmentation masks and performs preprocessing to generate standardized image representations. A hyperparameter optimization module generates candidate configurations for a deep learning segmentation model and evaluates the configurations using segmentation performance and domain-specific constraints. The domain constraints comprise one or more of boundary consistency, spatial continuity, region connectivity, anatomical preservation, segmentation smoothness, and tumor-region plausibility. An adaptive evolutionary mechanism dynamically updates the search process according to fitness improvement, convergence, population diversity, and constraint satisfaction. An interpretability module generates visual or feature-based representations indicating regions contributing to the segmentation output. A model selection module selects an optimized configuration based on segmentation performance, domain-constraint satisfaction, and interpretability criteria. The selected configuration is used to train a deep learning segmentation model for generating tumor segmentation maps from brain MRI images. The invention provides an integrated framework for adaptive optimization, domain-constrained segmentation, and interpretable brain tumor image analysis.
Disclaimer: Curated by HT Syndication.
