An Artificial Intelligence Generated Automated Algorithm to Measure Total Kidney Volume in ADPKD.

Area of Science:

• Nephrology
• Medical Imaging
• Artificial Intelligence

Background:

• Accurate prognosis tools for autosomal dominant polycystic kidney disease (ADPKD) are currently limited.
• Total kidney volume (TKV) is a key indicator for ADPKD progression and prognosis.
• Manual segmentation of kidney volume from MRI scans is time-consuming and subject to inter-observer variability.

Purpose of the Study:

• To develop and validate an artificial intelligence (AI)-based method for automated and routine measurement of TKV in ADPKD patients.
• To assess the performance of the AI tool compared to manual segmentation by human experts.

Main Methods:

• An ensemble U-net algorithm was developed using the nnUNet framework.
• The AI model was trained and internally validated on 1.5T MRI data from the CYSTic consortium (454 kidneys, 227 scans), with manual segmentation by a single operator.
• Independent validation was performed on 48 clinical MRI scans segmented by 6 different analysts.

Main Results:

• The AI algorithm achieved a median DICE score of 0.96 for both left and right kidneys, with a median TKV error of -1.8% on clinical validation data.
• Automated segmentation significantly reduced processing time, from 56 (±28) minutes for manual segmentation to 8.5 (±9.2) minutes for AI-assisted correction.
• The AI tool demonstrated high precision and accuracy comparable to manual segmentation.

Conclusions:

• The developed AI-based algorithm provides accurate and efficient TKV measurements in ADPKD patients.
• The tool's performance in real-world clinical settings indicates its suitability for routine application.
• This AI method has the potential to improve individual prognosis assessment in ADPKD.