A Lightweight SlimNet-Based Framework for Musculoskeletal Abnormality Detection

The increasing demand for musculoskeletal radio-graphic investigation significantly burdens radiologists, leading to delayed diagnosis times and increased risks of missed anomalies. This paper presents a comparative analysis between two lightweight deep learning models, SlimNet and EdgeNet, for the detection of musculoskeletal abnormalities using the Muscoskeletal Radiographs (MURA) dataset. SlimNet leverages depthwise separable convolutions integrated with Histogram of Oriented Gradients (HOG) features, while EdgeNet emphasizes edge-based feature extraction through learned convolutional filters. Both models are optimized to minimize computational overhead while achieving robust classification accuracy, enabling real-time diagnostic capabilities in clinical settings. We list the dataset preprocessing methods, architectural designs, and training techniques used to enhance model efficiency, sensitivity, and specificity. Experimental results demonstrate that EdgeNet consistently outperforms SlimNet + HOG in terms of accuracy, recall, and area under the Receiver Operating Characteristic curve (AUC), indicating superior generalization and stability. The comparative evaluation underscores EdgeNet’s effectiveness as a fast and reliable automated diagnostic tool, aiming to improve workflow efficiency in radiological practice, particularly in resource-constrained settings.