Efficient Thorax Disease Classification and Localization Using DCNN and Chest X-ray Images.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Thorax diseases, often bacterial lung infections, are life-threatening and necessitate timely diagnosis.
• Early detection of thoracic diseases is crucial for effective treatment and patient outcomes.
• Current diagnostic methods can be time-consuming, impacting rapid screening.

Purpose of the Study:

• To develop an automated system for swift and accurate detection and localization of thorax diseases using chest X-ray images.
• To enhance the capabilities of radiologists in diagnosing thoracic disorders.
• To create a precise computer-aided diagnosis (CAD) system leveraging deep learning.

Main Methods:

• Utilized the DenseNet-121 architecture as the foundation for the proposed Z-Net framework.
• Implemented a weighted cross-entropy loss function (W-CEL) to address class imbalance in the ChestX-ray14 dataset.
• Trained the model on 112,120 chest X-ray images for classification and localization tasks.

Main Results:

• The Z-Net framework achieved a mean Area Under the Curve (AUC) score of 85.8%.
• Demonstrated superior performance compared to previous models in detecting and localizing thorax diseases.
• Achieved the highest documented accuracy in the literature for related thorax disease detection models.

Conclusions:

• The proposed Z-Net model offers a highly accurate and precise CAD system for thorax disease diagnosis.
• The deep learning approach significantly improves the efficiency and accuracy of identifying thoracic diseases from X-ray images.
• This advancement supports faster patient screening and aids radiologists in clinical decision-making.