Home - Science & Technology - An AI model to assist the origin of a patient’s cancer.

An AI model to assist the origin of a patient’s cancer.

Share Post

An AI model to assist the origin of a patient's cancer.
An AI model to assist the origin of a patient's cancer.

Doctors are unable to pinpoint the cancer’s origin for a small percentage of cancer patients. Choosing a course of treatment for those people becomes significantly more challenging as a result, as many cancer medications are frequently created for particular cancer types.

A novel method created by MIT and Dana-Farber Cancer Institute researchers may make it simpler to pinpoint the places of origin for those perplexing tumors. The researchers developed a computational model that uses machine learning to examine the sequencing of roughly 400 genes and identify the location of a tumor’s origin in the body.

In a dataset of roughly 900 patients, the researchers demonstrated that they could reliably categorize at least 40% of tumors of uncertain origin with high confidence using this algorithm. Based on where their cancer began, this method allowed for a 2.2-fold increase in the number of individuals who would have been qualified for a genomically guided, targeted treatment.

Intae Moon, a graduate student in electrical engineering and computer science at MIT, is the study’s lead author. “That was the most important finding in our paper, that this model could potentially be used to aid treatment decisions, guiding doctors toward personalized treatments for patients with cancers of unknown primary origin,” she says.

The paper’s senior author, Alexander Gusev, is an associate professor of medicine at Dana-Farber Cancer Institute and Harvard Medical School. Nature Medicine published the paper today.

In 3 to 5 percent of cancer patients, particularly when tumors have spread to other parts of the body, physicians are unable to pinpoint the cancer’s original location. The designation “cancers of unknown primary” (CUP) applies to these malignancies.

This ignorance frequently limits medical professionals from being able to provide “precision” medications to patients, which are normally approved for particular cancer types where they are known to be effective. Compared to therapies intended to treat a wide range of malignancies, which are frequently administered to CUP patients, these focused treatments frequently work better and have fewer adverse effects.

“A significant number of people develop these cancers of unknown primary every year, and they have very limited treatment options because most therapies are approved in a site-specific way, where you have to know the primary site to deploy them,” Gusev explains.

In order to determine whether it was possible to forecast the type of cancer, Moon, a member of the Computer Science and Artificial Intelligence Laboratory who is also co-advised by Gusev, opted to examine genetic data that is routinely collected at Dana-Farber. The information consists approximately 400 or more genes’ genetic sequences, many of which are frequently altered in cancer. Using information from over 30,000 patients who had been identified as having one of the 22 recognized cancer types, the researchers created a machine-learning model.

The model was then evaluated on roughly 7,000 cancers whose origin site was known but which had never been observed before. The model, which the researchers called OncoNPC, had an accuracy rate of roughly 80% in predicting their genesis. Its accuracy increased to around 95% for tumors with high-confidence forecasts, which made up about 65 percent of the total.

Following these positive findings, the scientists used the model to examine a collection of roughly 900 tumors from CUP patients that were all obtained from Dana-Farber. They discovered that the model could produce high-confidence predictions for 40% of these cancers.

When combined with data from a subset of tumors, a study of the germline mutations, or inherited mutations, can reveal if a patient has a genetic predisposition to acquire a specific form of cancer. The researchers then compared the model’s predictions to the results of this analysis. The model’s predictions were considerably more likely to match the type of cancer that the germline mutations most strongly predicted than any other type of cancer, the researchers discovered.


Share Article