Automatic Evaluation of Histological Prognostic Factors Using Two Consecutive Convolutional Neural Networks on Kidney Samples

Abstract

Background and Objectives: The prognosis of patients undergoing kidney tumor resection or kidney donation is linked to many histological criteria. These criteria notably include glomerular density, glomerular volume, vascular luminal stenosis, and severity of interstitial fibrosis/tubular atrophy. Automated measurements through a Deep Learning approach could save time and provide more precise data. This work aimed to develop a free tool to automatically obtain kidney histological prognostic features. Design, setting, participants, and measurements: Two hundred and forty one samples of healthy kidney tissue were split into 3 independent cohorts. The "Training" cohort (n=65) was used to train two Convolutional Neural Networks: one to detect the cortex and a second one to segment the kidney structures. The "Test" cohort (n=50) assessed their performances by comparing manually outlined regions of interest to predicted ones. The "Application" cohort (n=126) compared prognostic histological data obtained manually or through the algorithm based on the combination of the two Convolutional Neural Networks. Results: In the Test cohort, the networks isolated the cortex and segmented the elements of interest with good performances (more than 90% of the cortex, healthy tubules, glomeruli, and even globally sclerotic glomeruli were detected). In the Application cohort, the expected and predicted prognostic data were significantly correlated. The correlation coefficients r were respectively 0.85 for glomerular volume, 0.51 for glomerular density, 0.75 for interstitial fibrosis, 0.71 for tubular atrophy, and 0.73 for vascular intimal thickness. The algorithm had a good ability to predict significant (> 25%) tubular atrophy and interstitial fibrosis level (ROC curve with an area under the curve 0.92 and 0.91, respectively) or a significant vascular luminal stenosis (> 50%) (area under the curve 0.85). Conclusion: This freely available tool enables the automated segmentation of kidney tissue to obtain prognostic histological data in a fast, objective, reliable and reproducible way.

Automated Computer-Assisted Image Analysis for the Fast Quantification of Kidney Fibrosis

“…Recently, a computerized approach to image analysis has been gaining prominence, new methods are being developed, and some are more sophisticated than others. Although there are better approaches such as artificial intelligence [ 17 ] and machine learning [ 18 ] that use convolutional neural networks trained to learn how to interpret the results of a given sample, reproducing the pathologist visual assessment [ 19 , 20 , 21 , 22 ], there are also more straightforward approaches such as image processing software. One of the most used is the free software ImageJ [ 13 , 23 , 24 , 25 , 26 , 27 ] and, more effectively, automated software such as CellProfiler™.…”

Artificial Intelligence-Assisted Renal Pathology: Advances and Prospects

“…With this model, the analysis results were not only close to the pathologist-determined scores of IF and TA but also significantly associated with patient outcomes. Recently, with advanced algorithms, the accuracy of AI in recognition of the finer structure and subtle changes in kidney biopsies has been improved, and the prediction power of allograft function has also been strengthened [ 82 , 83 ]. Thus, apart from the scoring of fibrosis extent, AI technologies may play a more crucial role in the prognosis and monitoring of post-transplant patients.…”

Artificial intelligence in renal pathology: Current status and future

“…Bouteldja et al developed a DL algorithm for accurate multiclass segmentation of digital WSIs of PAS-stained kidneys from multiple experimental renal pathology models (e.g., unilateral ureteral obstruction, ischaemia reperfusion injury, and adenine models) of various species, including mice, rats, pigs, marmosets, and bears [ 54 ]. Most recently, the prognostic features of kidney histology were automatically obtained by segmentation of the subcomponents of glomerular volume, glomerular density, interstitial fibrosis, tubular atrophy, and vascular intimal thickness [ 55 ]. Another recent approach is to detect and quantify interstitial, tubular, and mononuclear leukocyte infiltration in both preimplantation and posttransplant biopsies for risk stratification of allografts and posttransplantation monitoring in clinical practice [ 56 ].…”