Image-analysis algorithm to determine quality of cold perfusion in kidney transplantation

ABSTRACT NUMBER: NESS PRIZE FOR JUNIOR TRAINEES (BELOW ST3)_7

MAIN ABSTRACT TEXT

Introduction
Surgeon assessment of visual ‘quality of perfusion’ (QOP) influences kidney discard and predicts transplant outcome. However, this assessment is subjective and bias-prone. We aimed to design an “App” utilising a smartphone camera to make this assessment objective and enhance decision making.

Methods
The QOP in photographs of backbench kidneys was graded from 1 (ideal) to 5 (very poor) by three independent surgeons. A training cohort was used to develop an image-analysis algorithm, which was validated in a separate cohort.

Results
Analysing surgeon scores of 174 kidney images revealed that inter-rater agreement was good for kidneys displaying the best (rated 1) and worst (rated 4 or 5) QOP. However, for intermediate scores inter-rater agreement was poor. Inter-rater agreement between surgeons decreased as they graded more images; as surgeons fatigued, their ability to classify images worsened. A training cohort (n=174 kidneys) was used for algorithm development. First, small regions within each image were mapped within the ‘Red-Green-Blue’ colour-space, where well-perfused and poorly-perfused areas show clear separation. To generate a score for each kidney these regions are compared with ideally flushed kidney tissue. Testing our algorithm (validation cohort – n=29 kidneys) revealed strong correlation between image-analysis QOP score and surgeon assessment; r=0.789 (0.587-0.899), P<0.001.

Conclusion
Surgeon inter-rater agreement on kidney QOP is low for kidneys with borderline QOP, and worsens with fatigue. We provide a QOP score utilising an image-analysis algorithm, which correlates with surgeon scoring. With additional images and training this could provide an objective, numerical, point-of-care assessment of organ quality.