A Improved Training Method for Deep Learning Based Anatomical Classification of X-Rays

Every day, numerous medical images are sent to hospitals and tele-radiology facilities. These comprise a variety of techniques and body parts, such as elbow X-rays, head CT scans, and chest MRIs. These should be given to radiologists who are knowledgeable about the particular anatomy and modality. However, sometimes the scan's technicians enter the incorrect anatomy, employ a different naming scheme, or enter no information at all. There is a need for a vision-based anatomy classification system that can be applied in the absence or corruption of image-related meta-data. Before allocating a case to a radiologist, a manual screen may correctly estimate the anatomy from an X-Ray image. Our proposed solution will provide an automatic anatomy classification with improved training method using deep neural networks. We set out to develop a pre-trained supervised deep learning model that examines X-rays and returns the X-ray's anatomy from a preset set of anatomies in order to solve the given challenge. Here, we also suggest an improved training method to increase classification accuracy. With our present approach, we could accurately map reports and obtain anatomical classification accuracy of more than 95% using X-ray clinical data.