The BIOCHIP project: A Deep Learning approach for multiwell segmentation

The "Intelligent BIOsensors based on CHImeric Proteins"(BIOCHIP) project aims to design Machine Learning (ML) models to detect the presence of mercury (Hg2+) in water. Indeed mercury is dangerous for humans, even at low levels. If breathed, it can be lethal, and if absorbed through the skin, it can be toxic. Nowadays, such detection is performed by using expensive machines and instruments also requiring specialized personnel for a long time. To reduce time and allow non-specialized personnel to detect mercury, this research work presents ML methodologies to detect mercury through collected UV well plate images. The entire workflow is composed of two steps. The first step consists in developing a Neural Network (NN) able to recognize the ROI, i.e. the wells containing water to analyze, from the provided UV images. By using masks of these images provided by the BIOCHIP project, the NN is able to learn and consequently localize wells through a semantic segmentation task. Finally, as a second step, generated data have been used to predict the class to which each image belongs and also the respective concentration of mercury.