Optimal deep learning network for musculoskeletal X-Ray imaging classification for hospital medical record system

Background: Medical image archiving is one of the integral components of any hospital medical record system (HMRS). It includes, but is not limited to, MRI, CT-Scan, X-ray, Ultrasound, Musculoskeletal X-rays etc. The musculoskeletal X-ray images are relatively significant in number among the other types of medical imaging. Most of the existing HMRS use either the manual annotation of the images or use metadata for every image for archiving. This approach is found to be deficient because of intensive manual work, chances of misclassification, and reliance on human expertise. Moreover, archiving the images and their metafiles is relatively difficult to handle. Methodology: This issue can be handled by a hybrid solution of computer vision and deep learning. In the recent literature, researchers have proposed using machine learning and deep learning algorithms for biomedical image classification and archiving. However, the literature is found to be insufficient to recommend a unified deep learning network for Musculoskeletal X-ray Image classification with greater accuracy and efficiency. The LERA dataset is considered one of the benchmark Musculoskeletal X-rays image datasets. Results: To the best of knowledge the investigation of the best candidate of deep neural network is still missing in the literature. This study will present the logical and empirical rationale for the recommendation of the optimal deep learning network for X-ray Imaging Classification for Hospital Medical Record System using LERA (musculoskeletal radiographs) dataset. It has been concluded that the variants of Resnet, Google Net, and DarkNet are the suggested candidates for LERA x-ray image classification.