Deep learning quantification of percent steatosis in donor liver biopsy frozen sections

Abstract

Background Pathologist evaluation of donor liver biopsies provides information for accepting or discarding potential donor livers. Due to the urgent nature of the decision process, this is regularly performed using frozen sectioning at the time of biopsy. The percent steatosis in a donor liver biopsy correlates with transplant outcome, however there is significant inter- and intra-observer variability in quantifying steatosis, compounded by frozen section artifact. We hypothesized that a deep learning model could identify and quantify steatosis in donor liver biopsies. Methods We developed a deep learning convolutional neural network that generates a steatosis probability map from an input whole slide image (WSI) of a hematoxylin and eosin-stained frozen section, and subsequently calculates the percent steatosis. Ninety-six WSI of frozen donor liver sections from our transplant pathology service were annotated for steatosis and used to train ( n = 30 WSI) and test ( n = 66 WSI) the deep learning model. Findings The model had good correlation and agreement with the annotation in both the training set (r of 0.88, intraclass correlation coefficient [ICC] of 0.88) and novel input test sets ( r = 0.85 and ICC=0.85). These measurements were superior to the estimates of the on-service pathologist at the time of initial evaluation ( r = 0.52 and ICC=0.52 for the training set, and r = 0.74 and ICC=0.72 for the test set). Interpretation Use of this deep learning algorithm could be incorporated into routine pathology workflows for fast, accurate, and reproducible donor liver evaluation. Funding Mid-America Transplant Society

AI finally provides augmented intelligence to liver surgeons

“…In their paper untitled “Deep Learning Quantification of Percent Steatosis in Donor Liver Biopsy Frozen Sections » [1] , L. Sun et al. propose a new method to quantify graft macrosteatosis (MS) during liver procurement.…”

Efficiency of Machine Learning Algorithms for the Determination of Macrovesicular Steatosis in Frozen Sections Stained with Sudan to Evaluate the Quality of the Graft in Liver Transplantation

“…H & E is usually the standard procedure for general evaluation, which is easy and quick to perform and usually provides a good contrast to evaluate many parameters used to establish the quality of the graft for transplant. Nevertheless, this procedure does not stain fat, and the possibility of overestimating ME due to artifacts produced during the processing of the frozen biopsies (e.g., water droplets, holes, and so on) may be not discarded [ 3 , 10 ]. Taking this into account, coupled with the fact that ME determination is strongly observation-dependent, we find that the risk of error of judgement can increase significantly, with severe consequences, regardless of the final decision.…”

“…Nevertheless, this criteria is based on subjective evaluation by an experienced pathologist, which is strongly observation-dependent, non-reproducible, and challenging, even in experienced hands [ 8 , 9 ]. Additionally, the main limitation of the use of H&E frozen-stained sections is the risk of the underestimation of ME, due to the presence of artifacts (e.g., water droplets) during the sampling procedure [ 3 , 10 ]. Thus, the development of alternative technical and analytic procedures for staining and examining representative frozen graft sections, which allow for the establishment of an objective ME value in the shortest possible time, should be key to ensuring the viability of the transplant.…”

“…Nevertheless, the main limitation of all of these applications is that their analytic procedures are based on H&E stained sections, which may induce an underestimation of the value [ 3 , 15 ]. Although several deep learning segmentation algorithms [ 10 , 16 , 17 ] and supervised machine learning procedures [ 18 ] have been proposed to improve the accuracy of such algorithms in classifying intracellular fat vacuoles—identified as white spaces in H&E sections—the use of these applications is still not fully standardized. Other solutions have been proposed, such as the assessment of liver steatosis by liver texture analysis, a macroscopic determination which uses machine learning to speed up and automate the process [ 19 ]; a non-microscopic parameter which is still under validation.…”

State of machine and deep learning in histopathological applications in digestive diseases

“…Sun et al [ 60 ] took a modified VGG-16 patch-based segmentation approach to quantify macrovesicular steatosis in HE-stained frozen, donor liver biopsies. The network was trained on patches extracted from WSIs with steatosis regions annotated by pathologists.…”