Knowledge-Based Analysis for Mortality Prediction From CT Images.

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Healthcare
• Cardiovascular Disease Research

Background:

• Low-Dose Computed Tomography (LDCT) enhances lung cancer diagnosis accuracy over chest X-rays.
• Lung cancer screening populations face elevated risks for other critical conditions, including cardiovascular diseases.
• Accurate prediction of all-cause mortality in this demographic is essential for proactive healthcare management.

Purpose of the Study:

• To introduce a novel knowledge-based analytical method for predicting all-cause mortality.
• To develop a deep convolutional neural network (CNN) approach integrating imaging and clinical data.
• To assess the efficacy of combining quantitative clinical measurements with CNNs for mortality risk prediction.

Main Methods:

• Development of the Knowledge-based Analysis of Mortality Prediction Network (KAMP-Net).
• Extraction of structural image features from LDCT volumes at various scales using CNNs.
• Integration of quantitative clinical measurements and anatomical information with CNN-derived imaging features.

Main Results:

• Demonstrated the feasibility of using quantitative clinical measurements to enhance CNN-based mortality prediction from LDCT.
• Confirmed that radiologist-defined features significantly complement CNN performance.
• KAMP-Net achieved superior performance compared to existing methods in all-cause mortality prediction.

Conclusions:

• The KAMP-Net framework effectively integrates imaging and clinical data for robust mortality risk assessment.
• Incorporating quantitative clinical knowledge improves the predictive power of deep learning models in LDCT analysis.
• This approach offers a promising tool for improving patient outcomes in lung cancer screening programs.