Predicting dark-field images of H and E-stained esophageal specimens

The potential of laser-induced thermal therapy can be reassessed in treating abnormal mucosal tissues with advances in fiber optics, diode laser technology, and optical imaging modalities. In this context, studies optimizing a large parameter matrix (e.g., laser power, surface scanning speed, beam diameter, and irradiation duration) may be of interest. This study presents an artificial intelligence algorithm utilizing a generative adversarial network that predicts dark-field microscopy images from bright-field images of H&E-stained esophageal specimens. The calculated structural similarity index measurement between ground truth and the predicted dark-field image reaches an average of 74%. Also, the mean squared error is 0.7%.