A Hybrid Model for Abnormalities Detection in Upper Extremity Radiographs

Treating injuries by manual diagnosis of bone fractures in the upper extremities that are shoulder, humerus, elbow, wrist, finger, hand and forearm is a challenging task that can be prone to errors due to factors such as radiologist fatigue and experience level. For an accurate diagnosis and prompt treatment, anomalies of the upper extremities must be automatically detected in medical X-ray pictures. This research uses the MURA dataset to present a hybrid deep learning technique for multiclass categorization of anomalies in the upper extremity, encompassing the wrist, humerus, and elbow. This study uses pretrained Convolutional Neural Networks (CNNs) such as Visual Geometry Group (VGG19), Dense Convolutional Network (DenseNet121), and Residual Network (ResNet50), as well as customized CNN architectures, to classify abnormalities in upper extremity radiographs. These models are tailored to the anomaly detection job using transfer learning, resulting in accurate abnormality identification and localization. Furthermore, heatmap visualization techniques are used to improve anomaly identification outcomes, giving clinicians interpretable data for analyzing probable abnormalities in X-ray pictures.