Automated Classification Using End-to-End Deep Learning.

Area of Science:

• Medical Imaging and Diagnostics
• Artificial Intelligence in Healthcare
• Pulmonology

Background:

• Respiratory diseases like pneumonia and COPD are leading causes of death and hospitalization.
• Misdiagnosis rates for conditions such as COPD are alarmingly high (nearly 55%).
• Early diagnosis of lung diseases is critical for effective treatment and improved patient quality of life.

Purpose of the Study:

• To develop an End-to-End Deep Learning Lung Classification Model for Chest X-Ray (CXR) images.
• To classify CXR images into 14 primary classes of lung diseases.
• To enhance existing model accuracy for lung disease classification.

Main Methods:

• Implementation of a Densely Connected Convolutional Neural Network (DenseNet) architecture.
• Utilizing Deep Neural Networks for Computer Assisted Diagnosis (CAD) of lung conditions.
• Iteratively reducing search space and region of interest within CXR images.

Main Results:

• The study aims to improve upon existing lung disease classification accuracy using DenseNets.
• Performance comparison between a 14-class classification model and a binary classifier.
• Evaluation of DenseNet performance on CXR data with optimized search spaces.

Conclusions:

• Deep learning models, particularly DenseNets, show promise for accurate lung disease classification from CXR.
• Reducing the search space in CXR analysis can potentially improve diagnostic accuracy.
• Further research is needed to validate these findings and integrate AI into clinical practice for respiratory disease diagnosis.