BONE ABNORMALITIES DETECTION USING MEDICAL IMAGING
Keywords:
Bone Tumor and Fracture Detection (BTFC), X-ray Imaging (XRI), Deep Learning (DL), Convolutional Neural Networks (CNN), Explainable Artificial Intelligence (XAI), Bone Abnormalities (BA).Abstract
The early detection of any abnormalities in the bone, such as fracture and tumor, is of great importance for the timely clinical intervention. Lastly, the conventional diagnosis by X-ray is prone to errors, particularly in the cases where the image quality is poor, and there are only slight abnormalities. In this paper, we explain the OrthoVision system an automated system for bone abnormality detection and classification based on artificial intelligence for X-ray images, which will be implemented on a web-based basis. The system is able to classify into 4 classes fracture, non-affected, tumor-benign and tumor-malignant. It is based on EfficientNet-B0 and ResNet-18 ensemble model, along with CLAHE enhancement, resizing and converting to and normalizing a tensor. Visual explanation of model predictions, using GradCAM++, can be used to aid clinical interpretation. The validation performance reflects reliable predictions and generalizations, addressing the trustworthiness of the AI's accuracy and its potential usefulness for AI in real-world bone diagnostic tasks. The validation outcomes confirm effective predicative and generalizing performance, further emphasizing the usability of AI in bone diagnostics. The validations highlight trustworthy predictions and generalizations, underscoring the potential of using AI in bone diagnostics.
