In the search to find a cure for Diabetes, Cognexa teamed up with IKEM to develop a program that can automatically evaluate the quality of pancreatic islets. Pancreatic islets, or Islets of Langerhans, are clusters of cells containing beta cells which are responsible for the production of the hormone insulin. Beta cells are essential in regulating blood glucose and maintaining normal blood sugar levels. In diabetic patients, the immune system misidentifies beta cells as dangerous and destroys them, which then causes type 1 diabetes. In the hope of curing the disease, doctors are transplanting islet cells to patients
Identification of pancreatic islets among other types of tissue and the estimation of their total mass is crucial for choosing the right donors and acceptors for islet transplantation. Accurate assessment of the islet mass is also vital for medical research in a variety of pancreas-related pursuits, beyond the search for a cure for diabetes. This process, however, is cumbersome, time-consuming and prone to human error.
We developed a deep-learning neuron network built and trained in our cxflow, that segments and quickly identifies cells viable for transplant. After uploading batches of images to the website www.islenet.com, our program scans, analyzes and evaluates the images and promptly generates a report which doctors can download. For unlimited easy access, permanent links to uploaded batch collections can be generated upon request.
IsletNet is able to
segment the images with an over 99.2% f1 score, which provides accurate estimations
of islets counts and volumes (>0.99 R^2) in a matter of seconds. Our application
is conveniently accessible through our web application and is free of charge.
IsleNet accelerates and standardizes the evaluation process of pancreatic islets for surgical transplantation
1.5 months to MVP and 1.5 months to 8 full deployment and production
Segments images with a 99.2% f1 score, speeds up the process (10 seconds vs. 10 minutes), highest accuracy among all available tools (3% error-rate)