Image Processing in Toxicology: A Systematic Review

Cellular imaging has proven to be the key to identifying and applying biomarkers for tracking cell fate and drug activity in vitro. The aim here is to improve the comprehension of the mechanism of action of newly developed drugs. High-throughput imaging investigations, which quantify cellular morphological changes on a large scale, have become an invaluable instrument in the early phases of drug development due to automated microscopy and image processing. In order to document the evolution of cell morphology, photomicrographs have been taken using multi-well assay plates equipped with high-throughput imaging devices. These systems generate a multitude of results that take a lot of effort and subjective interpretation. Designing a step-by-step approach for detecting toxicity in assay images requires the automatic detection and classification of thousands of assay images. This method consists of acquiring images using high-throughput microscopy, processing the images, and analyzing the results. Because of the diversity of nuclei organization patterns associated with drug-induced toxicity, Deep Learning (DL) approaches have only just entered the landscape. These techniques have the significant advantage of screening thousands of images at different time intervals and at different spatial resolutions which would not be possible using manual intervention. We present a systematic review of the different image analysis techniques for toxicity detection in cell assay images. These include background correction, segmentation, and classification techniques. In addition, the review also focuses on the most prominent applications of DL to classify cell assay images as healthy or toxicity-affected. The advantage of DL based techniques is the flexibility of being incorporated in the high-throughput imaging pipeline along with the screening of thousands of assay images.