Deep learning approach to predict lymph node metastasis directly from primary tumor histology in prostate cancer.

Researchers

Journal

Modalities

Models

Abstract

To develop a new digital biomarker based on the analysis of primary tumor tissue by a convolutional neural network (CNN) to predict lymph node metastasis (LNM) in a cohort matched for already established risk factors.
Hematoxylin-eosin (HE) stained primary tumor slides from 218 patients (102 N+; 116 N0), matched for Gleason score, tumor size, venous invasion, perineural invasion and age, who underwent radical prostatectomy were selected to train a CNN and evaluate its ability to predict lymph node status.
With 10 models trained with the same data, a mean area under the receiver operating characteristic curve (AUROC) of 0.68 (95%-CI: 0.678-0.682) and a mean balanced accuracy of 61.37% (95%-CI: 60.05-62.69%) was achieved. Mean sensitivity and specificity were 53.09% (95%-CI: 49.77-56.41%) and 69.65% (95%-CI: 68.21-71.1%), respectively. These results were confirmed via cross-validation. The probability score for LNM prediction was significantly higher on image sections from N+ samples (mean N+ probability score: 0.58±0.17 vs. N0 probability score: 0.47±0.15, p=0.0016). In multivariable analysis, the probability score of the CNN (Odds ratio 1.04 per percent probability, 95%-CI: 1.02 – 1.08, p=0.04) and lymphovascular invasion (OR 11.73, 95%-CI: 3.96 – 35.7, p<0.0001) proved to be independent predictors for LNM.
In our experiment, CNN-based image analyses showed promising results as a potential novel low-cost method to extract relevant prognostic information directly from HE histology to predict the lymph node status of prostate cancer patients. Our ubiquitously available technique might contribute to an improved lymph node status prediction.
This article is protected by copyright. All rights reserved.