Many histologic parameters influence the prognosis of patients having kidney tumor excision or kidney donation. Among these criteria are glomerular density, glomerular volume, vascular luminal stenosis, and the degree of interstitial fibrosis/tubular atrophy. Deep-learning-based automated measures might save time and deliver more exact results. For a study, researchers sought to provide a free tool for automatically obtaining kidney histology prognostic characteristics. In all, 241 healthy kidney tissue samples were divided into 3 separate cohorts. The “Training” cohort (n=65) was used to train two convolutional neural networks, one to recognize cortex and the other to segregate kidney structures. The performance of the “Test” cohort (n=50) was evaluated by comparing manually delineated regions of interest to projected ones. The “Application” cohort (n=126) compared prognostic histologic data produced manually to data obtained using an algorithm based on a combination of two convolutional neural networks.
The networks performed well in the test cohort, isolating the cortex and segmenting the parts of interest (>90% of the cortex, healthy tubules, glomeruli, and even worldwide sclerotic glomeruli were found). The anticipated and predicted prognostic results in the Application cohort were strongly associated. The correlation coefficients r 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, respectively, were 0.85, 0.51, 0.75, 0.71, and 0.73 for vascular intimal thickness. Considerable tubular atrophy and interstitial fibrosis levels (receiver operator characteristic curves with area under the curves of 0.92 and 0.91, respectively) or significant vascular luminal stenosis (>50%) were well predicted by the method (area under the curve, 0.85). The technology automated kidney tissue segmentation in order to acquire prognostic histologic data in a quick, objective, reliable, and repeatable manner.