AI Business is part of the Informa Tech Division of Informa PLC

This site is operated by a business or businesses owned by Informa PLC and all copyright resides with them. Informa PLC's registered office is 5 Howick Place, London SW1P 1WG. Registered in England and Wales. Number 3099067.

Health & Pharma

Identifying hard-to-diagnose cancers with AI

by
 
Article ImageStudy: Data from large cell atlases can help profile tumors for targeted treatment

A machine learning model developed by MIT and other researchers could identify hard-to-diagnose cancers.

The Koch Institute for Integrative Cancer Research at MIT and Massachusetts General Hospital (MGH) implemented deep learning to evaluate genes in very specific ways to find the best treatment plan for patients. Gene expressions found in early cell differentiation and development could be the key to identifying mysterious cancers. The findings were published in Cancer Discovery.

“Sometimes you can apply all the tools that pathologists have to offer, and you are still left without an answer. Machine learning tools like this one could empower oncologists to choose more effective treatments and give more guidance to their patients,” said the study’s lead author Dr. Salil Garg, a pathologist at MGH and clinical investigator at the Koch Institute.

When a patient is first diagnosed with cancer, oncologists attempt to pinpoint where the cancer originated. If the clinician is unable to determine the starting point, the doctors start with non-targeted therapies, which often yield low patient survival rates.

Developing a machine learning model that factors in the differences between normal and healthy cells, along with the characteristics of various forms of cancers, can be challenging. A complex model with an overload of cancer gene expression features can stumble when taking in new data, even if it looks like it is learning the new data. Oversimplification of the model could lead to inaccurate diagnoses of cancer types.

Dr. Garg and his team narrowed in on signs of altered developmental pathways in tumor cells. They used the data from two robust cell atlases to identify correlations between embryonic and cancer cells. The Mouse Organogenesis Cell Atlas looked at 56 distinct trajectories of embryonic cells. The Cancer Genome Atlas profiled 33 types of tumors.

The correlative data was then used in their machine learning model called the Developmental Multilayer Perceptron (D-MLP), which analyzed the tumor and predicted its origin. The algorithm was used on 52 new samples. The platform was able to categorize the cancer types into four categories, which aided in the diagnosis and patient treatment plans.

In the future, the scientists plan on using additional data from tumor imaging, microscopy, and radiology to further enhance the predictive model for personalized medical treatments for cancer patients.

Trending Stories
All Upcoming Events

Upcoming Webinars

More Webinars

Latest Videos

More videos

EBooks

More EBooks

Research Reports

More Research Reports
AI Knowledge Hub

Newsletter Sign Up


Sign Up