The recent move by former NASA research technologist Thomas Fuchs shows that the interpretation of biopsies is somehow similar to mapping Mars’ rocky terrain. He has been utilizing some of the knowledge he gathered in training algorithms to navigate Mars in the Mars Rover project to create algorithms that can detect cancer.
According to Fuchs, his startup, Paige.AI, can utilize machine-learning in pathology, which can, in turn, save pathologists’ time, increase accuracy and provide better patient results. Pathology plays a fundamental role in the treatment of cancer patients. This step influences everything ranging from the ordered therapy or treatment to the extra tests required by doctors.
Paige. AI is concentrating on training algorithms for both breast and prostate cancer diagnosis. In February this year, the New York-based startup announced the securing of $25 million in a Series A round of financing that was led by Breyer Capital.
By utilizing digitized slides, Paige.AI applies deep learning methods to train algorithms, especially on how to differentiate between abnormal and healthy tissues. While using algorithms to identify the slides that were possibly cancerous, Fuchs asserted that pathologists would be in a position to spend a large portion of their time on analyzing results as opposed to sorting slides. He added that the algorithms have the potential to reduce the time pathologists spend sorting slides by half.
Fuchs, who plays the role of Paige.AI’s chief scientific officer and founder, also doubles as the director of computational pathology in the Warren Alpert Center for Digital and Computational Pathology located at Memorial Sloan Kettering Cancer Center. According to him, the institution is a popular resource, particularly for second opinions in pathology judging from the fact that nearly 30% of all cases handled by the center come from outside.
The recent agreement between New York-based Memorial Sloan Kettering and Paige.AI provides the startup with exclusive access to the institution’s intellectual property in computational pathology. Fuchs’ startup will also gain the opportunity to utilize the center’s library, which is made up of 25 million pathology slides, for the coming eight years.
Fuchs hopes that by perfecting the startup’s algorithms and digitizing its slides, pathologists could utilize Paige. AI’s pattern recognition capabilities to carry out an image search. Since raw images possessed by the Memorial Sloan Kettering library come with interpretations and genome sequencing tests, the image search could reveal the accompanying diagnosis, cure and the result of patients who were previously presented with the same cell morphology.
The genome sequencing data that accompanies an image, which matches a slide obtained from a fresh biopsy, can be utilized in predicting how the tissue could mutate. As such, a hospital might only require carrying out a single costly test to infer the outcomes of the second.
Training Paige. AI’s deep neural networks focus on a combination of several approaches. One of the techniques involves pathology fellows interpreting hundreds of cancer slides and then passing them to senior pathologists for review before feeding them to Paige. AI. The other method integrates the company’s software into the workflow of active pathologists.