Comparing Habitat, Radiomics, and Fusion Models for Predicting Micropapillary/Solid Components in Stage I Lung...

Area of Science:

• Oncology
• Radiology
• Medical Imaging

Background:

• Lung adenocarcinoma (LAC) is a major cause of cancer mortality.
• Accurate preoperative prediction of tumor subtype, such as micropapillary/solid (MP/S) status, is crucial for treatment planning in stage I LAC.

Purpose of the Study:

• To compare the efficacy of habitat models (HM), radiomics models (RM), and fusion models (FM) for predicting MP/S status in stage I LAC.
• To develop and validate advanced predictive models using imaging features.

Main Methods:

• Retrospective analysis of 345 stage I LAC patients from two centers.
• Development of HM using K-means clustering and RM using CT images.
• Creation of feature-based pre-fusion (pre-FM) and decision-based post-fusion (post-FM) models.
• Evaluation of model performance using Area Under the Curve (AUC) and Integrated Discrimination Improvement (IDI).

Main Results:

• The habitat model showed superior performance over the radiomics model in the training cohort (AUC: 0.900 vs. 0.876).
• The pre-fusion model consistently outperformed both HM and RM across all cohorts.
• The post-fusion model achieved the highest AUCs (0.952 in training, 0.850 in internal validation) and significant IDI improvements (14.2%-36.4%).

Conclusions:

• The post-fusion model, integrating clustering signatures, radiomics signatures, and clinical characteristics, is a reliable predictor for MP/S status in stage I LAC.
• This fusion approach offers enhanced preoperative prediction capabilities for lung adenocarcinoma subtypes.