MUMBAI, India, May 1 -- Intellectual Property India has published a patent application (202641048849 A) filed by Murugan Mageswari; Pavan Dhanasekar; Rithik Gv; Muniyappan R; Mohammed Sulaiman N; and Dr. Balamurugan A M, Chennai, Tamil Nadu, on April 16, for 'deep learning framework for accurate bone fracture classification from x-ray images.'
Inventor(s) include Murugan Mageswari; Pavan Dhanasekar; Rithik GV; Muniyappan R; Mohammed Sulaiman N; and Dr. Balamurugan A M.
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: "Inaccurate and delayed detection of bone fractures can significantly hinder timely and effective patient treatment, leading to complications and prolonged recovery times. However, manual interpretation of X-ray images remains a labor-intensive process that requires specialized expertise and is susceptible to human error, especially under heavy clinical workloads. To address these challenges, this study presents a deep learning-based framework for automated bone fracture detection designed to assist radiologists and orthopedic practitioners in improving diagnostic accuracy and efficiency. The proposed system employs a custom convolutional neural network (CNN) architecture enhanced with residual learning blocks to improve feature extraction and representation of subtle fracture patterns within X-ray imagery. Unlike traditional pre-trained models such as VGG16 and MobileNet, which often struggle with overfitting or insufficient generalization on medical image datasets, the residual connections in the proposed network facilitate deeper model training by mitigating vanishing gradient issues and enhancing the learning of fine-grained structural details. A curated dataset of 8,923 labeled X-ray images, encompassing multiple anatomical regions (including arm, leg, wrist, and ankle bones), was utilized for training, validation, and testing. Images were preprocessed through intensity normalization, contrast enhancement, and data augmentation techniques such as rotation and flipping to increase robustness and reduce class imbalance. The network was trained using a categorical cross-entropy loss function with an Adam optimizer, and hyperparameters were optimized through grid search to achieve optimal performance. Experimental results demonstrate that the proposed model achieves an accuracy of 84%, outperforming VGG16 (74%) and MobileNet (72%) on the same dataset. Furthermore, the model exhibits higher precision (0.86), recall (0.83), and F1-score (0.84) in identifying fractured regions, confirming its robustness in real-world diagnostic scenarios. These findings indicate that integrating residual learning within a CNN framework substantially enhances the model's capacity to identify intricate fracture features that may be overlooked by conventional methods. Overall, this work contributes an effective and reliable clinical decision support system that can complement radiological workflows, reduce diagnostic delays, and improve patient outcomes in orthopedic care. Future work will focus on extending the model to multi-class fracture localization and integrating explainable AI techniques to enhance interpretability and clinical trust."
Disclaimer: Curated by HT Syndication.
