A process that started several years ago is going to revolutionize the way we make tissue diagnosis. Digital pathology is the general name for high-resolution whole slide scanning generating digital images that can be viewed on the computer rather than using a microscope. The availability of a digital image allows for the application of artificial intelligence algorithms for identification of microscopic morphological events as well as prediction of clinical outcomes.

In our lab we apply a unique approach to generate effective algorithms to improve the diagnostic process. While deep learning methods are in general memory-based systems, our solutions combine logic and memory in intricate ways which enable us the development of solutions in domains which seems completely inaccessible to existing technologies. Hirschsprung’s disease is one such example.  Hirschprung’s disease is a classic example of a “needle in a haystack” task, where identification of isolated events of a specific cell type (ganglion cell) within a large tissue sample has critical implications on patients’ morbidity.   We chose to take this rare disease as a case study for our approach. In such case the total amount of data available in the world is too low to work in the standards the industry is expecting state of the art solution to produce meaningful results. We demonstrated that a small portion of the data available in one hospital enabled us to create a solution superior to human pathologists and to show that we can boost novice pathologist performance to be as good as expert level using less than 5% of time required for standards diagnosis. The solution we have created is robust, we tested the algorithm across different hospitals and scanners and achieved similar results.

The Hisprung`s solution was published in Scientific Reports (https://www.nature.com/articles/s41598-021-82869-y) and it is just one illustration of the type of solutions we are creating.