Optimized Ensemble Learning Technique on Wrist Radiographs using Deep Learning

Abstract

Radiographs of the musculoskeletal system provide significant expertise in the treatment of boned https://stanfordmlgroup.github.io/competitions/mura/isease (BD) or injury. To deal with such conditions Artificial Intelligence (Machine Learning & Deep Learning mainly) can play an important part in diagnosing anomalies in a musculoskeletal system. The approach in the proposed paper aims to create a more efficient computer diagnostics (CBD) model. During the initial stage of research, a few pre-processing techniques are used in the data set selected for wrist radiographs, which eliminates image size variability in radiographs. The given data set was then classified as abnormal or normal using three primary architectures: DenseNet201, Inception V3, and Inception ResNet V2. To improve performance of the model, the model's performance is then improved using ensemble approaches. The suggested approach is put to the test on a widely available MURA dataset also known as the musculoskeletal radiographs dataset, and the obtained outcomes are analyzed with respect to the reference document's current results. An accuracy of 86.49% was achieved for wrist radiographs. The results of the implementation show that the presented process is a worthy strategy for classifying diseases in bones.