Artificial intelligence-based method for renal function automatic assessment of each kidney using plain computed...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Nephrology

Background:

• Separate renal function assessment is crucial for clinical decisions.
• Current methods like SPECT are radioactive, time-consuming, and costly.

Purpose of the Study:

• To develop an automated method for assessing separate renal function using plain CT images and AI.
• To reduce radiation exposure, improve efficiency, and lower costs associated with renal function assessment.

Main Methods:

• A retrospective study involving 281 patients with nephratia or hydronephrosis.
• Deep learning (U-Net transformers) for automatic segmentation of renal and hydronephrosis regions.
• Radiomic feature extraction and modeling (ElasticNet, logistic regression) for function assessment.

Main Results:

• High accuracy in automatic kidney segmentation (mean DSC of 0.894 train, 0.881 test).
• Significant time savings: 3.4s (auto) vs. 1477.9s (manual) per case.
• The integrated AI model achieved an AUC of 0.825 in the test set for renal function assessment.

Conclusions:

• AI-powered analysis of plain CT scans is a feasible approach for assessing separate renal function.
• This AI method offers a promising alternative to SPECT, minimizing radiation risk and enhancing diagnostic efficiency.