Figure 9
(A) shows the same image before and after applying Eqs. (1) and (2) using \({{\omega }_{1}, \omega }_{2}\) provided in (C) for comparison. (B) Histograms of the same image before and after applying Eqs. (1) and (2). The application of Eq. (1) remarkably improves the image contrast, and (B) interpolates the input image according to \({{\omega }_{1}, \omega }_{2}\). PlexusNet-based models learn the optimal values \({{\omega }_{1}, \omega }_{2}\) during model training to determine the optimal nonlinear interpolation of the input image to solve a classification problem. The purpose of this approach is to contrast the semantic content of the input image to increase the likelihood of capturing meaningful features in the deep convolutional neural network layers that consequently impact the classification performance; \({{\omega }_{1}, \omega }_{2}\) given in (C) originated from the prostate cancer detection model. The example patch image has dimensions of ~ 512 µm × 512 µm at 10 × objective magnification. (C) Values of the input image were normalized using min–max normalization and correlated with the values from Eqs. (1) and (2).
