New AI Tool May Help Diagnose Neurodegenerative Diseases

Mount Sinai creates novel deep learning system for neurodegenerative diseases.

Posted Mar 07, 2019

geralt/pixabay
Source: geralt/pixabay

Recently, a team of led by pioneering researchers from Mount Sinai School of Medicine in New York City created one of the first platforms using large-scale image data in neuropathology for building and evaluating deep learning algorithms.

In a study published a few weeks ago in Nature’s Laboratory Investigation, the official journal of the United States and Canadian Academy of Pathology, Mount Sinai researchers developed a new deep learning algorithm using convolutional neural networks. The algorithm can recognize, classify, and quantify diagnostic elements of tauopathies—neurodegenerative disorders that may have glial or neuronal inclusions made of tau, a microtubule-binding protein.

The histopathological material used for the study was derived from twenty-two human autopsy brains from patients with tauopathies. Pathological tau in neurons form neurofibrillary tangles (NFT). The sections were digitized and uploaded to an informatics platform at Mount Sinai’s Center for Computational and Systems pathology. Over 80 million tests a year are processed at Mount Sinai's Department of Pathology, one of the largest academic pathology departments in the country with 62 full time pathologists and 900 histologists and laboratory technicians.

The convolutional network system was trained by the digitized images. The team deployed modified version of the fully convolutional SegNet architecture for the deep convolutional neural network generation, and used stochastic gradient descent for the differential loss function.

Interestingly, the update iterations were performed on “commodity GPU hardware” for efficient parallel processing. The neural network model was constructed using the PyTorch software package.

The researchers discovered a method using deep learning to improve tissue examination and supplement both partially quantitative and qualitative commonly-used techniques. As a result, the team’s new deep learning system provides a fast, reproducible, and unbiased method to “augment labor-intensive manual counting of histopathological features.”

Currently it is not known exactly how pathological tau impacts neuronal disorders. With this innovative framework, scientists have a way to access reproducible quantitative data for clinicopathological correlations to help in research of the pathogenesis of tauopathies such as Alzheimer’s disease.

Copyright © 2019 Cami Rosso All rights reserved.

References

Signaevsky, Maxim, Prastawa, Marcel, Farrell, Kurt, Tabish, Nabil, Baldwin, Elena, Han, Natalia, Lida, Megan A., Koll, John, Bryce, Clare, Purohit, Dushyant, Haroutunian, Vahram, McKee, Ann C., Stein, Thor D., White III, Charles L., Walker, Jamie, Richardson, Timothy E., Hanson, Russell, Donovan, Michael J., Cordon-Cardo, Carlos, Zeineh, Jack, Fernandez, Gerardo, Crary, John F..” Artificial intelligence in neuropathology: deep learning-based assessment of tauopathy.” Laboratory Investigation. 15 Feb 2019.

Irwin, David. “Tauopathies as Clinicopathological Entities.” Parkinsonism Related Disorders. Jan 22, 2016.

The Mount Sinai Hospital / Mount Sinai School of Medicine (2 Oct 2018). “Pathology test uses AI to predict prostate cancer progression following surgery.” Retrieved 3-7-2019 from https://www.eurekalert.org/pub_releases/2018-10/tmsh-ptu100218.php

More Posts