Habitat-radiomics combining multichannel 2.5D deep learning for differentiating adrenal adenomas from metastases...

Area of Science:

• Radiology
• Artificial Intelligence
• Oncology

Background:

• Qualitative diagnosis of lipid-poor adrenal adenomas and metastases is challenging.
• Accurate differentiation is crucial due to differing treatment principles and prognoses.

Purpose of the Study:

• To develop and validate a noninvasive model for differentiating lipid-poor adrenal adenomas from metastases.
• To improve diagnostic accuracy and guide precision treatment.

Main Methods:

• A total of 390 patients were included, divided into training, internal validation, and external test sets.
• A 2.5D deep learning model (DenseNet-121) and habitat-radiomics features were developed.
• A fusion model integrated deep learning scores, radiomics features, and clinical data, utilizing XGBoost for prediction.

Main Results:

• The fusion model achieved high diagnostic performance with AUCs of 0.983 (training), 0.913 (internal validation), and 0.886 (external test).
• Standalone deep learning and habitat-radiomics models also demonstrated strong predictive capabilities (AUCs 0.847-0.980 and 0.805-0.967, respectively).

Conclusions:

• The fusion model shows significant potential for noninvasive differentiation of lipid-poor adrenal adenomas and metastases.
• This approach can serve as a valuable tool for clinical decision-making and precision treatment planning.