In the medical field, billions of dollars are wasted each year — about $935 billion, but who's counting? According to a paper published by the JAMA Network, an estimated $75.7 billion to $101.2 billion is wasted through overtreatment. Of the many procedures that can lead to wasted resources, breast cancer biopsies are a major source of overtreatment. Houston Methodist Hospital is using artificial intelligence to create a more efficient and accurate Breast Cancer Risk Calculator, called iBrisk.
Breast cancer is something that plagues the lives of many women, and some men. According to the National Breast Cancer Foundation, one in eight women will be diagnosed with breast cancer in their lifetime.
Women are advised to start having annual mammograms to screen for breast cancer starting at age 40 to try to catch cancer in its earliest stages. With mammograms becoming a standard procedure, the process inevitably leads to more biopsies.
While more biopsies sound like the obvious course of action, Houston Methodist Hospital shares that out of 10,000 women biopsied, less than two will be positive while using the national standard. The result of a negative biopsy? Wasted time, resources, and money, as well as undue worry for the patient.
"It's not just wasteful. . .when you do an unnecessary procedure, you're potentially harming the patient," says Stephen Wong, Ph.D. After a negative biopsy, Dr. Wong explains that patients often begin to show emotional responses like high anxiety and low self-esteem. They often speculate the biopsies are wrong, and that they've had a missed cancer diagnosis by their medical provider.
Dr. Wong estimates that more than 700,000 patients have unnecessary biopsies in the breast cancer category alone.
Spearheading the iBrisk tool, Dr. Wong has found a way to utilize a smarter model than the current system for detecting breast cancer risk.
Hospitals across the country currently use the Breast Imaging Reporting and Database System score (BI-RADS), a system created by the American College of Radiology to determine breast cancer risk and biopsy decision-making.
To expand on BI-RADS data, Dr. Wong used multiple patient data points and AI technology to create the improved system. The iBRISK integrates natural language processing, medical image analysis, and deep learning on multi-modal BI-RADS patient data to make one of three recommendations: biopsy not recommended, consider biopsy, or biopsy recommended.
"While using AI, we try to simulate how the physician thinks," explains Dr. Wong. "The physician looks at different data: imaging, patient clinical data, demographic, history and other social factors. You don't rely on one particular thing."
To create iBrisk, Dr. Wong used 12 to 13 years of BI-RAD data at Houston Methodist Hospital to train the AI using deep learning.
He estimates that more than 80 percent of technical information is in the free text format, meaning unstructured data, in the United States.
"We applied an AI technique called natural language processing, which is using the computer to read the text automatically for us," explains Dr. Wong.
This data extraction tool was also used with imaging of mammogram ultrasounds by applying image analysis computer vision.
iBrisk also deploys deep learning, a machine learning tactic where artificial neural networks, inspired by the human brain, learn from large amounts of data. They determined approximately 100 parameters to analyze, including age, sex, socio-economic data, medical history, and insurance plans. After putting the data points into a deep learning method, the AI reduced the data points to the 20 risk indicators.
Houston Methodist Hospital used an estimated 11,000 cases for training, and then used 2,200 of its own data to test iBrisk. They have even been able to create unbiased independent validation by working with other hospitals like MD Anderson, testing their patients using iBrisk and confirming the results.
The potential of iBrisk to cut costs and contribute to less overtreatment has garnered support with other hospitals around the country. The breast cancer risk calculator is a collaboration with Dr. Jenny Chang of HMCC and breast oncologists at MD Anderson, UT San Antonio, and University of Utah Cancer Center.
While implicit racial bias has become a more prominent issue in the United States, Houston Methodist's iBrisk grants a neutral, unbiased lens. AI isn't immune to racial bias; in fact, computer scientist and founder of the Algorithmic Justice League, Joy Buolamwini, uncovered the large gender and racial biases of AI systems sold by IBM, Amazon and Microsoft in a 2019 article for Time.
With AI's history of racial bias in mind, Dr. Wong set out to create an impartial, fair system. "Our AI data is not sensitive to race. . .it's unbiased," he explains.
Houston Methodist Hospital plans to expand the iBrisk model to other forms of cancer in the future, including its next venture into thyroid and incidental lung nodule screenings.
The AI allows patients to save the stress of getting a biopsy.
"We are very careful to put any drugs or any procedure into clinical workflow until we are very sure you really have to pick this [outcome]," explains Dr. Wong. Using advanced risk detectors like iBrisk allows medical practitioners to make more thorough, informed decisions for patients looking into biopsies.
The categories are broken into low, moderate and high-risk groups. The low-risk groups have seen a 99.8 percent accuracy in results, missing only two cases out of a sample of 1,228. Patients that have fallen into the high-risk groups (leading patients to get a biopsy) have seen an 85.9 percent accuracy, compared to radiology, which is 25 percent accurate according to Dr. Wong.
Dr. Wong notes that patients that fall in the moderate section of the risk assessment can then have a dialogue with their physician to determine if they want to move forward with the biopsy. In the moderate category, there is a 93.4 percent accuracy.
If implemented, iBrisk would be able to reduce 75 percent of unnecessary biopsies, estimates Dr. Wong.
Currently, Houston Methodist Hospital is using AI technology outside of oncology, with the recent release of a tool that can diagnose strokes using a smartphone, announced in Science Daily. The tool, which can diagnose abnormalities in a patient's speech and facial muscular movements, was made in collaboration with Dr. Jay Volpi of Eddy Scullock Stroke Center at Houston Methodist Hospital.
"We are answering bigger questions," explains Dr. Wong, who looks forward to continuing to expand AI capabilities and risk calculators at Houston Methodist Hospital.
In the future, Dr. Wong looks forward to doing a multicenter trial to bring this technology outside of Texas.