A Robust Deep Learning Method with Uncertainty Estimation for the Pathological Classification of Renal Cell Carcinoma...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Accurate preoperative differentiation of renal cell carcinoma (RCC) subtypes is crucial for treatment planning.
• Computed tomography (CT) is a primary imaging modality for RCC evaluation.
• Deep learning offers potential for automated image analysis and classification.

Purpose of the Study:

• To develop and validate a deep learning model for preoperative classification of RCC subtypes (clear cell, papillary, chromophobe) using CT images.
• To incorporate uncertainty estimation into the model to quantify diagnostic confidence.
• To assist radiologists in making informed decisions for patients with RCC.

Main Methods:

• Retrospective collection of CT data from 668 RCC patients (Center 1).
• Development of a deep learning model using fivefold cross-validation for RCC subtype classification.
• External validation of the model using CT data from 78 patients (Center 2).
• Inclusion of uncertainty estimation in the model's predictions.

Main Results:

• The model achieved high performance in classifying RCC subtypes during cross-validation (AUCs ranging from 0.839 to 0.868).
• External validation demonstrated robust performance, with AUCs ranging from 0.787 to 0.856.
• Uncertainty estimation provided valuable confidence metrics alongside subtype predictions.

Conclusions:

• The deep learning model effectively differentiates RCC pathological subtypes based on CT imaging.
• Uncertainty estimation enhances the model's utility by providing diagnostic confidence.
• This approach supports improved preoperative decision-making for patients with renal cell carcinoma.