AI predicts breast cancer risk 5 years in advance: 3-year-old AI model goes viral after Anand Mahindra’s tweet

A study conducted in January 2021 by Massachusetts Institute of Technology’s CSAIL and Jameel Clinic on robust mammography-based models for breast cancer risk is now going viral after business tycoon Anand Mahindra recently shared a post on X and tweeted, “If this is accurate, AI is going to be more valuable to us than we imagine and much sooner than we imagine…(sic).”

We all normally have around 9,000 genetic mutations in our genome, making us genetic mutants to some degree. Although most of these mutations are benign and have little or no effect on our health, some can severely disrupt the function of proteins and cause diseases such as cancer. But what if artificial intelligence (AI) could accurately predict the formation of breast cancer years before it develops?

Studies have previously shown that we inherit most of these mutations from our ancestors, but accumulate about 64 mutations during our own evolution, passing them on to our offspring for good measure. And as we dream of having the best possible science at our fingertips, scientists at the Massachusetts Institute of Technology’s Computer Science and Artificial Intelligence Laboratory (CSAIL) and the Jameel Clinic for Machine Learning created a deep learning (DL) system in early 2021 that can predict cancer risk in women who have no signs or symptoms of breast cancer — just from breast X-rays or mammograms.

Viral AI breast cancer model

Designed to maintain consistent predictions despite slight clinical variations (such as different mammography machines), this AI system called “Mirai” can outperform existing risk-assessment algorithms because it predicts a patient’s risk at different future time points and can incorporate clinical risk factors such as age and family history, if available. According to a study published in Science Translational Medicine, Mirai was trained on more than 200,000 examinations from Massachusetts General Hospital (MGH), where it has now been installed and tested on patients placed outside MGH as well as the Karolinska Institute in Sweden and Chang Gung Memorial Hospital in Taiwan.

MIT’s AI system ‘Mirai’

Mirai not only maintained accuracy across different races, age groups, breast density categories, and cancer subtypes, but also outperformed the Tyrer-Cuzick model, identifying nearly twice as many future cancer diagnoses. Mirai consists of four modules:

1. An image aggregator module first collects and processes all conventional mammography images to generate a mammogram image.
2. It then collects image data from all scenes.
3. If necessary, the risk-factor prediction module uses mammography to predict a patient’s risk factors.
4. The final step involves an additive-hazard layer using the patient’s risk variables and mammography analysis to predict the patient’s risk annually for the next five years.

“Mammograms contain much more information than just four categories of breast density. Using deep learning models, we learn the subtle signals that indicate future cancer,” study lead author Adam Yalla from MIT in the US had said then.

MIT professor Regina Barzilay claimed, “Unlike traditional models, our deep learning model performs equally well across diverse ethnicities, ages, and family histories.”