Multi-head deep learning framework for pulmonary disease detection and severity scoring with modified progressive...

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Healthcare
• Computer Vision for Medical Diagnostics

Background:

• Chest X-rays (CXRs) are crucial for diagnosing pulmonary diseases, with billions performed annually.
• Early detection of conditions like COVID-19, pneumonia, and tuberculosis is vital for patient outcomes.
• Automated analysis of CXRs can assist radiologists in timely and accurate diagnoses.

Purpose of the Study:

• To develop a unified framework for pulmonary disease classification and severity scoring from CXRs.
• To segment lungs into six regions for detailed analysis and grading.
• To enable automated severity grading that correlates with the extent of opacity.

Main Methods:

• A novel framework integrating disease classification and severity scoring through lung segmentation.
• Implementation of a modified progressive learning technique with capped augmentations.
• Utilization of an internally generated attention map for efficient segmentation.
• Development of a single-digit severity score for four thoracic diseases.

Main Results:

• Achieved F1 scores of 0.924 (without fine-tuning) and 0.939 (with fine-tuning) on the BRAX dataset for segmentation and classification.
• Obtained a mean matching score of 80.8% for severity score grading.
• Attained an average area under the receiver operating characteristic curve of 0.88 for classification.
• Demonstrated competitive segmentation performance with significantly fewer parameters.

Conclusions:

• The proposed framework offers an effective approach for automated pulmonary disease classification and severity assessment using CXRs.
• The attention-based segmentation and modified progressive learning contribute to high performance and efficiency.
• This automated system has the potential to significantly aid radiologists in diagnosing and managing pulmonary conditions.