Background: An accurate assessment of graft weight and volume prior to liver transplantation is crucial to optimize organ allocation and ensure effective transplantation, especially in the paediatric transplantation setting. Our study aims to employ artificial intelligence (AI) protocols on donor CT scans to achieve highly precise predictions of these parameters.
Methods: We identified and analyzed morphological parameters and available CT images from 100 donors within our center in Turin over the past four years. For each donor, we ensured the availability of organ weight measured at the conclusion of bench surgery.
The data from 75 donors, along with total liver volumes obtained from their baseline CT scans, were used to train an AI software developed by our center in collaboration with the Polytechnic University of Turin. The software was subsequently employed to estimate the graft weight for the remaining 25 donors, and its results were compared with the known actual liver weight. Finally, the same technique was applied to evaluate the volume and weight of eight split grafts performed at our center, for which both CT and actual weight data were available.
In the secondary analysis, we combined CT data with biopsy results from all transplanted livers to stratify volume data based on the presence of steatosis. Additionally, donor morphological parameters were integrated to enhance the precision of weight and volume predictions.
Results: The error between the AI-estimated weight and the actual weight averaged 4% for whole livers and 6.5% for split grafts. The procedure can be considered as real-time, due to the rapidity of the process, estimated in 6-to 8 minutes.
Conclusion: The preliminary results of this study are promising . Increasing the number of cases to further train the AI model is expected to improve these outcomes.