Enhancing Lung Cancer Survival Prediction: 3D CNN Analysis of CT Images Using Novel GTV1-SliceNum Feature and PEN-BCE...

Area of Science:

• Oncology
• Artificial Intelligence
• Medical Imaging Analysis

Background:

• Lung cancer has a high mortality rate, with traditional survival analysis lacking accuracy due to subjectivity.
• There's a growing need for objective AI-driven survival analysis using clinical data and medical imaging.

Purpose of the Study:

• To enhance lung cancer patient survival classification using a 3D-CNN (ResNet-34) on CT images.
• To evaluate the impact of novel features and loss functions on classification performance.

Main Methods:

• Utilized the NSCLC-Radiomics dataset with a 3D-CNN (ResNet-34) architecture for CT image analysis.
• Conducted ablation studies to assess feature and methodology effectiveness.
• Introduced a novel feature (GTV1-SliceNum) and a novel loss function (PEN-BCE).

Main Results:

• Achieved a classification accuracy of 0.7434 and an ROC-AUC of 0.7768.
• Demonstrated superior performance compared to existing literature.
• Validated the efficacy of the novel feature and loss function in improving classification.

Conclusions:

• AI-driven survival prediction significantly improves lung cancer patient prognostication.
• The developed approach shows potential for enhancing personalized treatment strategies.
• Highlights the value of advanced AI techniques in oncological prognostic modeling.