MRI texture-based machine learning models for the evaluation of renal function on different segmentations: a...

Area of Science:

• Radiology
• Medical Imaging
• Machine Learning

Background:

• Assessing renal function noninvasively is crucial for managing kidney diseases.
• Current methods may be invasive or lack precision.
• MRI texture analysis offers a promising avenue for quantitative assessment of kidney health.

Purpose of the Study:

• To develop and validate a machine learning model using MRI texture features for noninvasive assessment of renal function.
• To compare different kidney segmentation methods for optimal model performance.

Main Methods:

• A retrospective study included 174 diabetic patients with varying degrees of renal function.
• Kidney segmentation was performed on T2-weighted MRI scans using four distinct methods.
• Texture features were extracted using Speeded-Up Robust Features (SURF) and a Support Vector Machine (SVM) model was constructed.
• Diagnostic performance was evaluated using Receiver Operating Characteristic (ROC) curve analysis.

Main Results:

• The models utilizing the largest coronal slices (LC-K) and all coronal slices (All-K) for segmentation demonstrated the highest accuracy.
• The optimal model achieved high diagnostic performance in classifying normal, mild/moderate, and severe renal impairment.
• Area Under the Curve (AUC) values in training cohorts ranged from 0.919 to 0.959, and in validation cohorts from 0.802 to 0.863.

Conclusions:

• An MRI-based machine learning model was successfully developed and internally validated for accurate renal function assessment.
• This noninvasive model has the potential to aid in monitoring patients with impaired renal function.
• External validation is recommended to confirm the model's generalizability and clinical utility.