Drop Volume Algorithm and Analysis System

Project Goal: Design, build and verify a system that uses light sensor data to store and calculate the volume of cancer-identifying reagents being dispensed onto a tissue sample, enabling technicians to confirm reagent volumes and avoid false diagnoses.

Roche Tissue Diagnostics’ automated tissue staining systems dispense set amounts of dye, or reagent, onto patient tissue samples to diagnose conditions such as cancer. To deliver the most precise and accurate diagnoses, the system needs to dispense the proper quantity of reagent on every sample. The team developed a new system that uses a neural network to calculate an accurate volume for each drop.

The system consists of a photoelectric sensor which is set in the path of the reagent dropper and sends a stream of analog input to a microcontroller. When a drop is detected, the microcontroller isolates the drop instance and performs preprocessing on it to prepare the analog input for the algorithm. The microcontroller then uses a pre-trained recurrent neural network to calculate an accurate volume for the drop. The volume calculation is saved with other bookkeeping data, then converted to analog and output to the Roche tissue staining system to verify the drop volume.

This system, built on advanced neural networks and made to integrate with the existing Roche Tissue Diagnostics tissue staining system, helps to verify the integrity of the tissue staining system at each drop, preventing patients from receiving false diagnoses.