Multiple instance learning for lung pathophysiological findings detection using CT scans.

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Medicine
• Pulmonology

Background:

• Lung diseases cause millions of premature deaths globally.
• Computed Tomography (CT) scans reveal characteristic pathological findings.
• Traditional Computer-Aided Diagnosis (CAD) methods analyze limited image regions, potentially missing crucial data.

Purpose of the Study:

• To apply Multiple Instance Learning (MIL) for comprehensive lung CT scan analysis.
• To detect and characterize various lung pathophysiological findings.
• To predict Epidermal Growth Factor Receptor (EGFR) mutation status in lung cancer.

Main Methods:

• Utilized Multiple Instance Learning (MIL) algorithms for image analysis.
• Focused on detecting specific lung pathologies: Fibrosis, Emphysema, Satellite Nodules, Nodules in Contralateral Lung, and Ground Glass.
• Applied MIL for predicting EGFR mutation status in lung cancer patients.

Main Results:

• Achieved an Area Under the Curve (AUC) of 0.89 for Fibrosis detection and 0.72 for Emphysema detection.
• Demonstrated an AUC of 0.69 for EGFR mutation status prediction.
• The MIL approach provided a comprehensive analysis of lung pathophysiological changes.

Conclusions:

• The MIL-based approach offers a robust method for characterizing lung diseases from CT scans.
• This comprehensive analysis aids in identifying key lung pathologies and predicting cancer-related markers.
• The study highlights the potential of MIL as a valuable tool in clinical decision-making for lung conditions.