Automated Femoral Head Segmentation Using Rcnn: a Robust Approach With Coco Dataset and Advanced Augmentation Techniques
Year: 2024
Type: article
Abstract: Segmentation of the femoral head is crucial in orthopedic diagnostics, facilitating the identification and management of diseases such femoroacetabular impingement, avascular necrosis, and osteoarthritis. This study presents an automated method for femoral head segmentation utilizing a Region-based Convolutional Neural Network (RCNN) model, applied to a dataset of 2,035 femoral head pictures. Pre-processing techniques, such as auto-orientation, scaling to 640x640 pixels, contrast enhancement using contrast stretching, greyscale conversion, and noise reduction, were utilized to standardize the input data. Data augmentation was executed through horizontal flipping, 90-degree rotations, and modifications to hue, saturation, brightness, blurring, and noise levels to enhance model robustness. The RCNN model underwent training for 300 epochs, attaining exceptional outcomes with a mean Average Precision (mAP) of 99.5%, precision of 99.7%, and recall of 100%, so illustrating the model's efficacy in precisely segmenting the femoral head. Notwithstanding the favorable results, the dataset's restricted size may hinder the model's generalizability to varied real-world applications. Future endeavours will concentrate on augmenting the dataset and incorporating multimodal vision methodologies to guarantee conformity with contemporary healthcare demands and norms. These advances will improve the system's clinical utility and offer a more thorough solution for orthopedic imaging and diagnostic processes. This research advances the field of medical picture segmentation by providing a scalable and efficient method for automated femoral head segmentation, with potential for further enhancement to satisfy real-time clinical requirements.
Authors: Syed Osama Ali Shah, Alishba Eman
Institution: Bahria University
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: bronze
OpenAlex ID: W4403776533