Bone Tumor Detection and Classification Using Fast Mask Region-Based Convolutional Neural Network

Abstract

Accurate and timely diagnosis of bone tumors is paramount for effective treatment and patient well-being. Leveraging medical imaging modalities such as radiographs and magnetic resonance imaging (MRI), we introduce a novel methodology for automated bone tumor detection and classification. Our approach centers on the utilization of the Fast Mask R-CNN (Region-based Convolutional Neural Network) architecture, renowned for its efficiency in object detection and segmentation tasks. The workflow begins with image preprocessing steps aimed at enhancing contrast and eliminating noise, crucial for optimal performance in subsequent stages. Subsequently, the Fast Mask R-CNN framework is deployed to detect and precisely delineate bone tumor regions within the images, effectively isolating them from surrounding anatomical structures. This segmentation facilitates accurate localization, a crucial step in the diagnostic process. Following tumor localization, a classification model is employed to categorize the identified regions into benign or malignant types, leveraging the distinctive radiological features characteristic of each. This classification task is accomplished using a convolutional neural network (CNN) trained on a curated dataset of annotated bone tumor images. By combining the strengths of Fast Mask R-CNN for precise localization and CNN for accurate classification, our methodology achieves enhanced accuracy and reliability in bone tumor detection and classification. This innovative approach holds significant promise in streamlining diagnostic workflows and improving patient outcomes in bone tumor management.