MRI-Based Grading of Clear Cell Renal Cell Carcinoma Using a Machine Learning Classifier

Area of Science:

• Radiology
• Machine Learning
• Oncology

Background:

• Clear cell renal cell carcinoma (ccRCC) grading is crucial for treatment decisions.
• Accurate differentiation between low-grade (ISUP I-II) and high-grade (ISUP III-IV) ccRCC is clinically significant.
• MRI texture analysis offers a non-invasive method for characterizing renal tumors.

Purpose of the Study:

• To develop and validate a machine learning classifier for distinguishing low-grade from high-grade ccRCC using MRI texture features.
• To assess the performance of the ML model in a retrospective patient cohort.
• To explore the utility of quantitative MRI texture analysis in ccRCC grading.

Main Methods:

• Retrospective analysis of 99 ccRCC patients (61 low-grade, 38 high-grade).
• Manual delineation of tumor regions of interest (ROIs) on CMP sequence MRI images.
• Extraction of quantitative texture features using MaZda software, including histograms, co-occurrence, run-length, gradient, and autoregressive models.
• Development of a multi-layer perceptron classifier trained on selected texture features and evaluated on a validation set.

Main Results:

• 257 texture features demonstrated high reproducibility (Intraclass Correlation Coefficient [ICC] ≥ 0.80).
• Six key features (Kurtosis, 135dr_RLNonUni, Horzl_GLevNonU, 135dr_GLevNonU, S(4,4)Entropy, S(0,5)SumEntrp) were selected for the classifier.
• The trained multi-layer perceptron model achieved 95.7% accuracy on the training set and 86.2% on the validation set.
• Area under the receiver operating curves (AUC) were 0.997 (training) and 0.758 (validation).

Conclusions:

• A machine learning-based grading model utilizing MRI texture analysis can effectively discriminate between low-grade and high-grade ccRCC.
• This ML model shows potential as an adjunct tool for clinical diagnosis and management of ccRCC.
• Further validation in larger, prospective cohorts is warranted to confirm its clinical utility.