Breast Cancer Early Detection Possible With New AI Technology

As many as 3.1 million American women are living with a history of breast cancer as of January 2019. The numbers are only rising as 268,600 new cases of invasive breast cancer and 62,930 new cases of non-invasive breast cancer are expected to be diagnosed in 2019.  

The statistical predictions are bad news for women in the U.S. as breast cancer is the second most deadly variety of cancer likely to turn fatal (for women), closely following lung cancer. Sadly, scientists still remain skeptical about early detection methods even as more than 41,760 deaths of women due to breast cancer are anticipated in 2019.

Regular screening of breasts done with mammography tests have not proven useful to stop the cancer in its nascent stages for women under the age of 40, studies have revealed. Though a handful of randomized clinical trials have shown mammography tests are capable of reducing mortality rates above the age of 50, the data is still not sufficient to conclude benefits of routine mammography tests and checking breast tissue density to prevent cancer from escalating in younger women.  

Current modules of diagnosis rely on a few indicators such as family history, ovarian cancer, hormonal and reproductive issues and most importantly, breast density. Since the correlations between these conditions are not very strong, there is a lack of accuracy in diagnosing breast cancer in the early stages for women younger than 40. False positive and false negative tests are thus given to women who are easily misled. 

Scientists from Massachusetts Institute of Technology (MIT)’s Computer Science and Artificial Intelligence Laboratory ( CSAIL) and Massachusetts General Hospital (MGH) have developed a new breakthrough with the use of artificial intelligence (AI) to predict future breast cancer risks. 

They created a pattern to identify subtle changes in the breast tissues that indicate high cancer risk and that are otherwise not identifiable in the current module of diagnosis used by clinicians in U.S, especially as manual identification is prone to human error. The scientists proved that having dense breast tissues is not a factor involved in diagnosing high risk of breast cancer and this criterion which is used popularly to gauge high risk of breast cancer does not hold true anymore.

“When our hybrid DL model was compared with breast density, we found that patients with nondense breasts and model-assessed high risk had 3.9 times the cancer incidence of patients with dense breasts and model-assessed low risk,” the researchers said in their study, which is published in the medical journal Radiology.

mit-breast-cancer-model-696x464 AI detects that woman is at high risk of breast cancer four years in advance. MIT CSAIL

The innovation came about by applying a hybrid mix of solutions such as assessing medical history and using full field mammography images. It was not restricted to manually pointing out the calcification in the breast tissues, hence this AI is touted to be a major advancement in medical technology. 

The artificial intelligence used here is capable of combining Deep Learning (DL) technology that not only takes full field mammograms, but it is also programmed to process data analysis to find subtle cues in the breast tissues, indicating cancerous formations within the next five years.   

“To distinguish a model’s ability to predict future cancer development from its ability to detect cancers on the basis of the current mammography, we compared models on a subgroup of the test set by excluding mammography from women in whom cancer was diagnosed in less than 3 years. We observed that our models showed similar performance when predicting future risk,” the researchers said.

In a press release published by MIT Computer Science and Artificial Intelligence Lab, this new deep learning module is said to be 31 percent more accurate for patients in the high risk category, while only 18 percent of accuracy could be traced in the old way of detection.

Out of 80,243 mammography examinations used for training and validation, 3.4 percent or 3,045 patients received a cancer diagnosis within five years. Similarly, out of 8,751 used for testing, 269 or 3.1 percent were diagnosed with cancer within a five-year period.

The researchers were categorical about the inclusion of African American females in the study and acknowledged they are more prone to breast cancer before the age of 40.