AI-based pipeline for early screening of lung cancer: integrating radiology, clinical, and genomics data

Area of Science:

• Oncology
• Radiology
• Artificial Intelligence

Background:

• Lung carcinoma prognosis improved with targeted therapies for molecular targets like Epidermal Growth Factor Receptor (EGFR) mutations.
• Next-generation sequencing for EGFR mutation detection is not widely accessible in resource-limited countries like India.
• Intratumoral heterogeneity and tissue adequacy challenges necessitate AI-driven solutions for lung nodule analysis and EGFR mutation prediction.

Purpose of the Study:

• To develop and evaluate an AI-based pipeline for automatic lung nodule detection and characterization from CT images.
• To predict the probability of EGFR mutations in lung carcinoma patients using AI models.
• To provide a cost-effective and non-invasive method for EGFR mutation status determination in resource-limited settings.

Main Methods:

• Utilized CT imaging and EGFR gene sequencing data from 2277 lung carcinoma patients across multiple cohorts.
• Trained an AI system (AIPS-Nodule or AIPS-N) for automatic lung nodule detection and feature prediction.
• Developed an AIPS-Mutation (AIPS-M) model combining AIPS-N results with clinical factors to predict EGFR genotype, evaluated using Area Under the Curve (AUC).

Main Results:

• AIPS-N demonstrated effective lung nodule detection with an average AP50 of 70.19% and predicted five lung nodule properties.
• The AIPS-M models (machine learning and deep learning) achieved AUCs ranging from 0.587 to 0.910 in predicting EGFR genotype.
• The AI system successfully identified patients with EGFR mutations.

Conclusions:

• CT imaging combined with an automated AI lung nodule analysis system can predict EGFR genotype.
• This AI-driven approach offers a cost-effective and non-invasive method for identifying EGFR mutations in lung cancer patients.
• The system aids oncologists and patients in resource-limited settings to achieve optimal care and therapy selection.