Artificial intelligence, or AI, has become a hot topic in countless industries. Whether it’s using AI to predict consumer habits and grow the economy, teach new languages, or streamline healthcare systems, the potential is endless for how machine learning can be applied to human life.
AI can be used in many facets of the healthcare industry. It can be used to process billing and administrative information, bring diagnostic care to underserved populations, and make medical equipment more effective. In addition, AI is making a splash in the realms of cancer detection, diagnosis, and treatment.
Detecting cancer early on, before it progresses, can be the difference between life or death for a patient and can save healthcare professionals time, money, and effort. A team of engineers at the University of Central Florida’s Center for Research in Computer Vision are currently learning to incorporate AI into their research and work.
The team designed an AI system trained to skim through CT scans and detect lung cancer tumors. After feeding the system 1,000 CT scans, they found that it could detect tumors with a 95 percent accuracy.
Similar success can be found around the world. Scientists at Showa University in Japan have created AI-based technology to detect colorectal cancer in early stages. Detecting cancers before they become malignant improves prognosis. Using their AI technology the Showa team is able to detect colorectal cancer with 86 percent accuracy.
These machines are the first step in eliminating cancers, because they stop the disease before it becomes malignant. Perfecting these systems has the potential to reduce treatment spending, patient stress, and loss of human life.
Additionally, AI is being used to treat cancer. Due to the severity of cancers, and human error, doctors can make overzealous diagnoses that result in false positives. Not only is this mentally and physically draining for the patient, but it wastes time.
Researchers at the Massachusetts Institute of Technology have designed an AI to address this problem in breast cancer detection. The MIT News reported that the team had great success in reducing the number of false positives in breast cancer patients.
Regina Barzilay, MIT’s Delta Electronics Professor of Electrical Engineering and Computer Science, and recent MacArthur genius grant recipient, wrote a report on the team’s findings, published October 2017 in the medical journal Radiology.
“Because diagnostic tools are so inexact,” said Barzilay, “there is an understandable tendency for doctors to over-screen for breast cancer. When there’s this much uncertainty in data, machine learning is exactly the tool that we need to improve detection and prevent over-treatment.”
The team’s AI uses machine learning to predict if lesions will need surgery through previously acquired data. The machine used scans of 600 existing high-risk lesions and looked for patterns in data like demographics, family history, and medical records to learn which high-risk lesions were likely to need surgery.
The first iteration of the test was successful. After training, it was given a new set of 335 high-risk lesions to test. Correctly diagnosing 97 percent of the cancerous lesions as malignant, and reducing the number of unnecessary surgeries by 30 percent.
MIT’s success is exciting for all kinds of cancer and proves the efficacy of AI. This past August Canion Medical Research Europe in Edinburgh, Scotland received a £140,000 grant to develop an AI machine to help diagnose malignant pleural mesothelioma.
For a disease like mesothelioma, AI capabilities could truly impact patients in a positive way. Caused by the carcinogenic mineral asbestos, the cancer is difficult to detect due to its long latency of 20 to 50 years. Detecting mesothelioma early could expand the life expectancy of these patients. According to Ken Sutherland, the president of Canion Medical Research Europe, they hope to use remaining grant money to improve cancer treatment.
“We believe that an automated assessment method using AI would be a major advance in fighting this disease,” Sutherland said.
Treatment planning for any disease is important, but can be a long and laborious task. For cancer, radiation and chemotherapy are the most common treatments, both of which are harmful to the patient even when they work.
Radiologists are tasked with aiming radiation at diseased sites, while trying to keep the healthy tissue around the tumor intact. Tumors that reside close to vital parts of the human body like the neck, heart, and lungs are often most difficult to treat in this way.
DeepMind Technologies, Google’s artificial intelligence and machine learning division, is developing a system to efficiently direct radiation. Using scans from patients who suffered from head and neck tumors, DeepMind is creating an algorithm to quantify the doctor’s decision making patterns to automate this part of treatment. DeepMind’s approach gives AI the ability to teach itself how to treat tumors, this hands-on approach to machine learning can be tested by using previous patients’ scans.
IBM is also tailoring cancer treatment using their Watson supercomputer. Watson takes a more historical approach. Using synthetic medical records and big data to develop treatment plans, Watson extrapolates from given data to recommend treatments.
However, STAT News reported in July 2018 that Watson made multiple “unsafe and incorrect treatment recommendations,” and still has difficulty distinguishing between different kinds of cancer. Perhaps this is because it’s been trained on synthetic records instead of real ones, or perhaps it’s because the technology simply isn’t there yet.
At this point, AI is most useful in the detection and diagnosis stage but hurtling towards breakthroughs in treatment as well.
As AI systems become more advanced in a number of disciplines, professionals may worry that they could suffer job loss due to machine advances. That fear is mostly unfounded. Though specialized AI technology performs well, more board AI machines tend to have less accuracy.
The thought of error-free doctors acting without human lapses in judgement or limitations is a thrilling prospect, but far away from reaching the general public. Narrow and well-trained AI technology may be more prevalent in general hospitals soon, however, an infallible machine doctor is not yet possible. Nevertheless, scientists and engineering teams work every day to make this promising technology a reality.