Interpretation of Chest x-rays affected by bullets using deep transfer learning

TL;DR

This study employs deep transfer learning to classify and localize bullet-affected areas in chest X-rays, demonstrating the model's effectiveness across various body parts and providing a novel approach in medical imaging diagnostics.

Contribution

It introduces the first deep learning-based method for detecting and classifying bullet injuries in radiographs, applicable to multiple body regions.

Findings

High accuracy in classifying bullet-affected X-rays

Effective localization of bullet impacts in images

Model generalizes across different body parts

Abstract

The potential of deep learning, especially in medical imaging, initiated astonishing results and improved the methodologies after every passing day. Deep learning in radiology provides the opportunity to classify, detect and segment different diseases automatically. In the proposed study, we worked on a non-trivial aspect of medical imaging where we classified and localized the X-Rays affected by bullets. We tested Images on different classification and localization models to get considerable accuracy. The replicated data set used in the study was replicated on different images of chest X-Rays. The proposed model worked not only on chest radiographs but other body organs X-rays like leg, abdomen, head, even the training dataset based on chest radiographs. Custom models have been used for classification and localization purposes after tuning parameters. Finally, the results of our findings manifested using different frameworks. This might assist the research enlightening towards this field. To the best of our knowledge, this is the first study on the detection and classification of radiographs affected by bullets using deep learning.