Predicting post-donation renal function loss by measuring renal cortex volume using AI
Development of an AI-based automatic segmentation model for measuring renal cortex volume using preoperative CT images
- A larger preoperative renal cortex volume correlates with less decline in renal function after donation... Useful as a predictive indicator
- Expected to be useful in predicting the prognosis of elderly donors... Supporting safer donation decisions
A recently developed AI-based renal cortex volume measurement model can easily and accurately predict renal function loss after kidney donation, according to a study. The model is expected to revolutionize the existing complex and time-consuming kidney evaluation method, helping elderly donors make safer donation decisions.
Based on data from 1,074 living donors who underwent kidney donation surgery from 2010 to 2020, Professor Min Sangil's team (Professor Jo Eun-Ah of Chung-Ang University Hospital, Professor Lee Juhan of Severance Hospital, and CEO of OncoSoft Kim Jin Sung) at the Department of Transplantation and Vascular Surgery at Seoul National University Hospital announced the results of a multi-institutional retrospective cohort study on August 26th, which measured renal cortical volume using AI-based CT images and analyzed the correlation between renal cortical volume and post-transplant renal function decline.
As we enter an aging society, kidney transplant surgeries from elderly donors are increasing. However, in the case of elderly donors, renal function is likely to decline due to changes in the renal microstructure, such as glomerular sclerosis due to aging. Therefore, a method to predict post-donation renal function loss became necessary. Previous studies have reported a correlation between renal cortical volume and renal function, but the existing measurement methods were complicated and time-consuming, making it difficult to apply in actual clinical practice.
The research team developed an "AI-based automatic segmentation model" to solve these problems. This AI model was designed to automatically measure the cortical volume of the kidney by analyzing pre-donation CT images. Since the renal cortex plays a vital role in kidney function, accurately measuring this volume is crucial for predicting renal function.
To validate the model's accuracy, The research team compared the cortical volume measured by the AI model with manual measurements. The verification results showed that the accuracy of the AI model was very high, recording a Dice similarity coefficient (an index that evaluates the overlapping part between two images, where a value closer to 1 indicates a higher level of similarity) of 0.97 and a Hausdorff distance (a measure of the maximum error between the predicted boundary and the actual boundary, where a smaller value indicates more accuracy) of 0.76 mm. This means that the AI model can measure the actual renal cortex volume very accurately.
The research team analyzed the correlation between renal cortical volume, measured using an AI model, and post-donation renal function (estimated glomerular filtration rate, eGFR). eGFR is an indicator of the kidney’s filtration capacity, and a lower value indicates a decline in renal function. They utilized the generalized additive model (GAM) to analyze renal function changes over time after the donation.
[Figure 1] The renal cortex drawn by the research team (left) and AI (right). The AI model can measure the actual renal cortex volume very accurately.
[Figure 2] Changes in renal function after kidney donation. Older donors (red) tend to show a greater decline in renal function in the first 1 to 3 years after donation compared to younger donors (blue).
As a result, older donors (over 60 years old) were found to tend to experience a greater decline in renal function after donation than younger donors (under 60 years old). Specifically, the decline in eGFR in older donors was statistically significantly greater (P = 0.041), demonstrating that older donors experienced more decline in renal function.
However, donors with larger preoperative renal cortex volume tended to have less renal function decline after donation. This difference was statistically significant (p<0.001), especially in older donors. This indicates that older donors with larger renal cortex volumes can maintain renal function better after donation.
The research team highlighted that using AI-based technology to measure the renal cortex volume could be a crucial indicator in predicting the loss of renal function after donation. They also noted that incorporating this model into the donor selection and evaluation process is expected to significantly improve kidney transplant surgery success rates and enhance donor safety.
Professor Min Sangil (Department of Transplant Vascular Surgery) said, "This study, which demonstrated the clinical benefits of using AI to measure renal cortical volume, is seen as a major advancement in assessing and predicting the prognosis of kidney donors." He also emphasized, "This is particularly significant as it can aid in making safer decisions regarding donations by offering a more accurate method to predict loss of renal function, especially in the case of elderly donors."
This study was published online in the International Journal of Surgery (IF=12.5), a world-renowned journal in the field of surgery.
[Pictures from left] Professor Min Sangil of the Department of Transplant Vascular Surgery at Seoul National University Hospital, Professor Cho Eun-Ah of the Department of Transplant Vascular Surgery at Chung-Ang University Hospital