MUMBAI, India, May 1 -- Intellectual Property India has published a patent application (202641051515 A) filed by Srinivasa Ramanujan Institute Of Technology; Dr D. Maruthi Kumar; M. Kusuma; U. Om Ganesh; and K. Rangaswamy, Ananthapuramu, Andhra Pradesh, on April 22, for 'seeded cnn refinement of k-means clustering for accurate mri brain tumor segmentation.'
Inventor(s) include Srinivasa Ramanujan Institute Technology; Dr D. Maruthi Kumar; M. Kusuma; U. Om Ganesh; and K. Rangaswamy.
The application for the patent was published on May 1, under issue no. 18/2026.
According to the abstract released by the Intellectual Property India: "Seeded CNN Refinement of K-Means Clustering for Accurate MRI Brain Tumor Segmentation Abstract: Accurate brain tumor segmentation from Magnetic Resonance Imaging (MRI) is a critical step in computer-aided diagnosis, supporting clinical decision-making for detection, treatment planning, and disease monitoring. Conventional clustering-based techniques such as K-Means are computationally efficient, but they often generate coarse tumor boundaries and produce false positive regions due to intensity inhomogeneity and complex tumor morphology. In contrast, Convolutional Neural Networks (CNNs) can achieve high segmentation accuracy; however, they typically require large annotated datasets and considerable computational resources, which may be difficult to obtain in medical imaging environments. To address these limitations, this paper proposes a hybrid pipeline, Seeded CNN Refinement of K-Means Clustering for Accurate MRI Brain Tumor Segmentation. In the proposed method, K-Means clustering is first applied to obtain an initial coarse segmentation of tumor candidate regions. These candidate masks are then used as seed regions for a patch-based CNN that refines boundary delineation, improves structural consistency, and suppresses false detections. By restricting CNN processing to clustering- guided candidate areas, the framework reduces computational overhead and lowers dependence on extensive labelled data. Experimental evaluation on benchmark MRI brain tumor datasets demonstrates that the hybrid approach outperforms standalone K-Means and CNN-only baselines across multiple metrics, including Dice Similarity Coefficient (DSC), Intersection over Union (IoU), sensitivity, specificity, precision, and 95% Hausdorff distance. The results indicate that the proposed method provides an effective balance between segmentation accuracy and computational efficiency, making it suitable for practical clinical applications."
Disclaimer: Curated by HT Syndication.
