Radiogenomics in Clear Cell Renal Cell Carcinoma: Machine Learning-Based High-Dimensional Quantitative CT Texture...

Area of Science:

• Radiology and Medical Imaging
• Oncology
• Machine Learning in Medicine

Background:

• Clear cell renal cell carcinoma (ccRCC) is the most common subtype of kidney cancer.
• PBRM1 gene mutations are frequent in ccRCC and are associated with prognosis.
• Accurate prediction of PBRM1 mutation status is crucial for guiding treatment decisions.

Purpose of the Study:

• To evaluate the efficacy of machine learning (ML)-based quantitative CT texture analysis for predicting PBRM1 gene mutation status in ccRCC patients.
• To assess the performance of artificial neural network (ANN) and random forest (RF) algorithms in this prediction task.

Main Methods:

• Retrospective analysis of CT images from 45 ccRCC patients (PBRM1 mutated and non-mutated).
• Data augmentation to create 161 labeled segmentations for model training and validation.
• Extraction of high-dimensional texture features from contrast-enhanced CT images, followed by ML model development (ANN, RF) and 10-fold cross-validation.

Main Results:

• 759 out of 828 extracted texture features demonstrated excellent reproducibility.
• The ANN algorithm achieved 88.2% classification accuracy (AUC 0.925) using 10 features.
• The RF algorithm achieved 95.0% classification accuracy (AUC 0.987) using 5 features, outperforming the ANN (p=0.007).

Conclusions:

• ML-based quantitative CT texture analysis is a feasible method for predicting PBRM1 mutation status in ccRCC.
• The Random Forest algorithm demonstrated high accuracy and potential for non-invasive PBRM1 mutation status prediction in ccRCC.
• This approach may aid in personalized treatment strategies for ccRCC patients.