To get cancer earlier, we need to predict who will get it in the future. The complex nature of the predictive risk has been amplified by artificial intelligence (AI), but the use of AI in medicine has been limited by poor performance for new patient populations and the failure of racial minorities.
Two years ago, a team of scientists from MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) and Jameel Clinic (J-Clinic) demonstrated a deep learning system to predict cancer risk using only a patient’s mammogram. The model showed significant promise and even improved inclusivity: it was equally accurate for both white and black women, which is especially important since black women are 43 percent more likely to die from breast cancer.
But to integrate image-based risk models into clinical care and make them widely available, the researchers say the models require algorithmic improvements and large-scale validation in multiple hospitals to prove their robustness.
To that end, they have adapted their new “Mirai” algorithm to capture the unique requirements of risk modeling. Mirai collectively models a patient’s risk over several future time points and may optionally benefit from clinical risk factors such as age or family history, if available. The algorithm is also designed to provide predictions that correspond to small deviations in clinical environments, such as the choice of a mammography machine.
The team trained Mirai in the same dataset of more than 200,000 Massachusetts General Hospital (MGH) exams from their previous work, validating it on test kits from MGH, the Karolinska Institute in Sweden, and Chang Gung Memorial Hospital in Taiwan. Mirai is now installed at MGH, and the team members are actively working on integrating the model into care.
Mirai was significantly more accurate than the previous methods of predicting cancer risk and identifying high-risk groups in all three data sets. When comparing high-risk cohorts on the MGH test set, the team found that their model identifies almost twice as many future cancer diagnoses compared to the current clinical standard, the Tyrer-Cuzick model. Mirai was also accurate in patients of different races, age groups and breast density categories in the MGH test set and in different cancer subtypes in the Karolinska test set.
“Improved breast cancer risk models enable targeted screening strategies to achieve earlier detection, and less screening damage than existing guidelines,” said Adam Yala, PhD student and lead author of CSAIL in an article on Mirai published this week in Science Translational Medicine. “Our goal is to make this progress part of the standard of care. We are working with clinicians from Novant Health in North Carolina, Emory in Georgia, Maccabi in Israel, TecSalud in Mexico, Apollo in India and Barretos in Brazil to further validate the model on different populations and study how to best clinically to implement. ”
How it works
Despite the widespread acceptance of breast cancer screening, the researchers say the practice is riddled with controversy: More aggressive screening strategies are aimed at maximizing the benefits of early detection, while less frequent screenings aim to false false, anxiety and cost for those reduce which will never even develop breast cancer.
Current clinical guidelines use risk models to determine which patients should be recommended for complementary imaging and MRI. Some guidelines use risk models with just age to determine if and how often a woman should be screened; others combine various factors related to age, hormones, genetics and breast density to determine further tests. Despite decades of effort, the accuracy of risk models used in clinical practice remains modest.
Recently, in-depth mammography-based risk models have shown promising performance. To bring this technology to the clinic, the team identified three innovations that they believe are critical to risk modeling: joint modeling time, the optional use of non-image risk factors, and methods to ensure consistent performance in clinical settings.
1. Time
Inherent in risk modeling is the learning of patients with different amounts of follow-up and the assessment of risk at different times: this can determine how often they are screened, whether they should have additional imaging or even consider preventative treatments.
Although it is possible to train separate models to assess the risk for each time point, this approach can lead to risk assessments that do not make sense – such as predicting that a patient is at greater risk of cancer within two years. developed as within five years. . To address this, the team designed their model to simultaneously predict risk, by naming an instrument that is an ‘additive hazard layer’.
The additive hazard layer works as follows: their network predicts the risk of a patient at a time, such as five years, as an extension of their risk at the previous time, such as four years. In this way, their model can learn from data with varying amounts of follow-up and then provide self-consistent risk assessments.
2. Non-image risk factors
Although this method focuses primarily on mammograms, the team also wanted to use non-imaging risk factors such as age and hormonal factors if available, but did not require them during the test. One approach is to add these factors as an input to the model with the image, but this design will prevent most hospitals (such as Karolinska and CGMH), which do not have this infrastructure, from using the model.
For Mirai to take advantage of risk factors without requiring it, the network predicts the information during the training period, and if it is not there, it can use its own predictive version. Mammograms are rich sources of health information, and so many traditional risk factors such as age and menopausal status can be easily predicted from their imaging. Because of this design, the same clinic can be used worldwide, and if they have the additional information, they can use it.
3. Constant performance in clinical settings
To incorporate deep-learning risk models into clinical guidelines, the models must perform consistently in different clinical settings, and their predictions cannot be influenced by small variations such as the machine on which the mammogram was taken. Even in a single hospital, scientists have found that standard training does not provide consistent predictions before and after a change in mammography machines, as the algorithm can learn to rely on different indications specific to the environment. To de-bias the model, the team has a opponent scheme where the model specifically learns mammogram representations that are immutable to the clinical environment, to provide consistent predictions.
To further test these updates in different clinical settings, the scientists evaluated Mirai on new test kits from Karolinska in Sweden and Chang Gung Memorial Hospital in Taiwan and found that it achieved consistent performance. The team also analyzed the model’s performance in breeds, ages, and breast density categories in the MGH test set and on cancer subtypes in the Karolinska data set, and found that it performed similarly in all subgroups.
“African American women continue to have breast cancer at a younger age, and often at a later stage,” said Salewai Oseni, a breast surgeon at Massachusetts General Hospital who was not involved in the work. ‘This, together with the higher incidence of triple negative breast cancer in this group, has led to an increased mortality rate for breast cancer. This study shows the development of a risk model whose prediction has significant accuracy in race. The opportunity for clinical use is great. ”
Here’s how Mirai works:
1. The mammogram image is inserted by something called an “image encoder”.
2. Each image representation, as well as the view from which it came, is combined with other images from other views to obtain a representation of the entire mammogram.
With the mammogram, the patient’s traditional risk factors are predicted using a Tyrer-Cuzick model (age, weight, hormonal factors). If it is not available, predicted values are used.
4. With this information, the additive hazard layer predicts a patient’s risk for each year during the next five years.
Mirai improves
Although the current model does not look at any of the patient’s previous imaging results, changes in imaging over time contain a wealth of information. In the future, the team aims to create methods that can effectively utilize the full image history of a patient.
In a similar way, the team notes that the model can be further improved by using ‘tomosynthesis’, an X-ray technique to examine asymptomatic cancer patients. In addition to improving accuracy, additional research is needed to determine how to adapt image-based risk models to different mammography devices with limited data.
“We know that MRI can detect cancer earlier than mammography, and that earlier detection improves patient outcomes,” says Yala. ‘But for patients at low risk for cancer, the risk of false positive benefits may outweigh. With improved risk models, we can design more nuanced risk screening guidelines that provide more sensitive screening, such as MRI, to patients who will develop cancer, to get better results, while reducing unnecessary screening and overtreatment for the rest. ”
“We are very excited and humbled to ask whether this AI system will work for African American populations,” said Judy Gichoya, MD, MS and Assistant Professor of Interventional Radiology and Informatics at Emory University, who is not involved. was at work. “We study this question in detail and how to detect failure.”
Yala co-authored the paper on Mirai with Peter G. Mikhael, MIT Researcher, Radiologist Fredrik Strand of Karolinska University Hospital, Gigin Lin of Chang Gung Memorial Hospital, Associate Professor Kevin Smith of KTH Royal Institute of Technology, Professor Yung-Liang Wan of Chang Gung. University, Leslie Lamb of MGH, Kevin Hughes of MGH, senior author and professor at Harvard Medical School Constance Lehman of MGH, and senior author and MIT professor Regina Barzilay.
The work was supported by grants from Susan G Komen, Breast Cancer Research Foundation, Quanta Computing and the MIT Jameel Clinic. It was also supported by Chang Gung Medical Foundation Grant and Stockholm Läns Landsting HMT Grant.