Automatic Sequence-Based Network for Lung Diseases Detection in Chest CT.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Existing AI systems often overlook bacterial pneumonia (BP) in CT screening, focusing primarily on viral pneumonia (VP).
• A unified approach is needed for differential diagnosis of VP, BP, and normal cases using chest CT volumes.

Purpose of the Study:

• To develop an accurate, rapid, and interpretable AI system for lung disease diagnosis using computed tomography (CT).
• To enable differential diagnosis between viral pneumonia, bacterial pneumonia, and normal cases.

Main Methods:

• A novel sequence-based pneumonia classification network (SLP-Net) was developed, treating CT volumes as image sequences.
• SLP-Net employs sequential Convolutional Neural Networks (CNNs) and Convolutional Long Short-Term Memory (ConvLSTM) for feature extraction and axial map construction.
• An adaptive-weighted cross-entropy loss (ACE) and sequence attention maps were utilized for optimization and enhanced classification confidence.

Main Results:

• SLP-Net demonstrated significant performance in classifying viral pneumonia, bacterial pneumonia, and normal cases on a dataset of 258 chest CT volumes.
• The proposed method outperformed existing slice-based and volume-based approaches, even with limited training data.
• The model achieved superior classification accuracy, highlighting its efficacy in differential diagnosis.

Conclusions:

• SLP-Net offers an effective and data-efficient AI solution for the differential diagnosis of lung diseases from CT scans.
• The sequence-based approach captures crucial spatial information, leading to improved diagnostic accuracy.
• This interpretable AI system has the potential to enhance clinical CT screening for various lung conditions.