M ³Lung-Sys: A Deep Learning System for Multi-Class Lung Pneumonia Screening From CT Imaging.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Accurate COVID-19 diagnosis is critical for pandemic control.
• Limited data and resources pose challenges for developing diagnostic tools.
• CT imaging is a key modality for detecting lung pneumonia.

Purpose of the Study:

• To develop a deep learning system for multi-class lung pneumonia screening from CT images.
• To address limitations of training data and resources in developing diagnostic AI.
• To enable accurate differentiation of COVID-19 from other pneumonias (H1N1, CAP) and healthy cases.

Main Methods:

• Proposed a Multi-task Multi-slice Deep Learning System (M³Lung-Sys) utilizing two 2D CNNs.
• Employed slice-level and patient-level classification networks for feature extraction and temporal information recovery.
• Developed lesion localization capabilities without requiring pixel-level annotations.

Main Results:

• M³Lung-Sys demonstrated superior performance in both slice- and patient-level classification tasks.
• The system achieved high accuracy in distinguishing COVID-19 from Healthy, H1N1, and CAP cases.
• Generated lesion location maps provided interpretability and clinical value.

Conclusions:

• M³Lung-Sys offers an effective and efficient solution for multi-class lung pneumonia screening using CT imaging.
• The model's ability to localize lesions enhances its clinical utility for diagnosis and treatment planning.
• This deep learning approach can aid in managing infectious respiratory disease outbreaks.