MicroNET: Microarray Image Orientation Predictor and Automated Data Extraction

Microarray is a high-throughput platform consisting of several thousands of DNA, protein, or peptides on a chip surface. In the current scenario, researchers have to use a scanner such as GenePix®4000B to scan such chips. However, after scanning these chips, researchers need to manually adjust the orientation angle of the grid to be placed in the microarray image before moving on to the spot intensity extraction step and data analysis. This paper introduces a novel lightweight deep-learning model, MicroNET, to predict the orientation angle and top left coordinates of the grid. After the image is corrected and the grid is placed, we also perform automatic fluorescence intensity extraction. This paper has compared two different configurations of MicroNET. The first one makes use of a ResNet18 backbone while the next one is a modified version of the ResNet18 backbone with lesser parameters. As per our results, the model with the modified ResNet18 backbone predicts the orientation angle with an MSE of 0.379 and the top left coordinates with an MSE of 0.0011. Thus the model with the modified ResNet18 backbone achieves better results with a lesser number of parameters. Hence it can be easily executed on any microarray data extraction software and will enable researchers to prepare microarray datasets easily.