AI Platform Shows Potential in Diagnosing Patients at Early Stages
Conducted in collaboration with The Michael J. Fox Foundation (MJFF), the study’s findings support the platform’s potential to aid in diagnosing Parkinson’s and allowing patients to begin treatment while still in the disease’s early stages.
It worked by detecting and analyzing alpha-synuclein protein aggregates in peripheral nerve tissue samples collected through a biopsy and also examined by trained pathologists.
“Objective diagnostic tools, especially early in disease, are critical to drive care decisions and to design trials toward better treatments and cures,” Jamie Eberling, PhD, senior vice president of research resources at the MJFF, said in a press release.
The study, “Antemortem detection of Parkinson’s disease pathology in peripheral biopsies using artificial intelligence,” was published in the journal Acta Neuropathologica Communications.
Parkinson’s disease is characterized by the buildup of toxic alpha-synuclein protein aggregates, called Lewy bodies, in the central nervous system or CNS (the brain and spinal cord), particularly in dopamine-producing neurons. However, this buildup is also seen in other neurodegenerative diseases, including Lewy body dementia (LBD), and multiple system atrophy (MSA).
Such overlap poses a challenge in diagnosing Parkinson’s early, since these diseases often share various motor and non-motor symptoms.
A definitive Parkinson’s diagnosis is currently achieved only with a post-mortem analysis of a patient’s brain tissue.
While alpha-synuclein aggregates accumulate mostly in the CNS, they are also present in the peripheral nerves — those that send sensory and motor information from the brain and spinal cord to the rest of the body.
In a previous study, researchers at the Icahn School of Medicine at Mount Sinai, in New York, showed that toxic alpha-synuclein aggregates can be visualized in biopsies of the peripheral nerves of patients at advanced disease stages. Specifically, they analyzed the peripheral nerves of the submandibular gland, one of two salivary glands located below the lower jaw.
AI-automated analysis would aid pathologists in decision-making and standardize how tissue samples are analyzed, a process that currently is time consuming and can be prone to error, as it relies on a person’s training and judgment.
Screening that relied on AI, however, was not optimal for early-stage Parkinson’s patients.
“Our previous results indicate that peripheral Lewy-type synucleinopathy (LTS) is present in early PD [Parkinson’s disease], suggesting its utility as a diagnostic and prognostic biomarker,” the researchers wrote, “however, the sensitivity was moderate.”
These same scientists now used PreciseDx’s AI-based digital pathology technology to analyze 95 submandibular gland biopsies collected from patients enrolled in the Systemic Synuclein Sampling Study or S4 (NCT02572713), an MJFF initiative.
S4 aimed to compare the levels of alpha-synuclein in multiple fluids — saliva, blood, and cerebrospinal fluid — and in tissues, including those of the submandibular gland, across patients at different disease stages and within the same patient.
The PreciseDx platform is powered by the company’s AI Morphology Feature Array, an algorithm that is able to quantify and assess an array of features specific to a certain disease from a single tissue slide.
“Traditionally, pathology grading systems look at a few morphology components to make a diagnosis. Unlike any human-powered grading method, PreciseDx’s AI Morphology Feature Array (MFA) can examine thousands of different features and leverage those relationships between them,” said John F. Crary, MD, PhD, a professor of pathology and molecular and cell-based medicine at the Icahn School.
The AI platform, initially developed for breast cancer with MFA used to assess patient risk and enable more personalized treatment, was now tested for its potential with Parkinson’s disease.
Initially, the platform was trained to recognize and identify alpha-synuclein aggregates in peripheral nerve salivary gland biopsy samples taken from early-stage Parkinson’s patients. Training relied on the annotations of three expert pathologists, who analyzed the samples in a blind manner, meaning they did not know the status of the person they were taken from.
During the training, PreciseDx’s AI-platform detected features of Parkinson’s disease in the biopsy samples with a sensitivity and specificity of 99% compared with the experts’ diagnosis. Of note, a test’s sensitivity is its ability to correctly identify those with a given disease, while specificity is its capacity to correctly identify those without it.
Following training, researchers assessed the platform’s performance in a separate set of 1,230 tissue slides taken from 42 Parkinson’s patients and 14 healthy adults enrolled in the S4 study. The analysis was conducted using features assessed as highly specific for Parkinson’s.
PreciseDx’s platform was now seen to have a sensitivity of 71% and a specificity of 65% in predicting clinical Parkinson’s status, and a 64% accuracy in predicting Parkinson’s stage. It outperformed the expert neuropathologists’ results for sensitivity (59%) but not specificity (88%).
“Our study demonstrates that deep machine learning represents a feasible way to augment routine histological [biopsy] examination, and trained neural networks could be deployed in detecting the peripheral LTS and improving accuracy for further confirmatory assessment by a neuropathologist,” the researchers concluded.
“This industry-changing study has shown that we need to revitalize the way we think about pathology and lean into using AI to detect diseases more accurately,” Crary said.
The researchers noted that further work is needed to “develop more sensitive and specific tests,” adding “we will continue [to] broaden and improve our research protocol and approach to the application of AI in neuropathology.”
“We look forward to working with PreciseDx as it explores the potential of utilizing the AI platform in pathology across multiple diseases, including Parkinson’s,” said Erik Lium, PhD, president of Mount Sinai Innovation Partners, and executive vice president and chief commercial innovation officer for the Mount Sinai Health System.