Artificial Intelligence (AI) shows significant potential in medical technology. Computers are already diagnosing Alzheimer’s, Pneumonia, and eye diseases. The future of medicine looks hopeful even in the area of cancer. But this claim would require vast amounts of patient data from over the years to deem plausible. This data fed to computers could promise insights that can transform health care.
Cancer is a complex disease. The word cancer is devious in its homogenous sense. Each diagnosis is different, and each type of cancer has multiple approaches to treatment. Drug responses vary for different individuals. AI can provide more accurate predictions with these large data sets and the speed of discovery for a cure will depend on the rate of its availability. However, an immediate substantial change is not likely. But with eventual development and adoption of analytics processes in healthcare companies, it can transform cancer care in areas of diagnosis and treatment.
AI in Diagnostic Tests
Identifying and treating cancer would secure better survival rates if the diagnosis was accurate. And AI has proved useful in reducing the rate of errors in diagnosis. Early diagnosis in developed nations is highly probable, with many featuring screening guidelines for breast cancer. But the ultimate goal of algorithms today is to take over the tedious jobs, so we have the time to pursue other areas of interest.
If funding and research began for screening and diagnostic tests, it is still rare to find patients with all the data that is needed to proceed. There has to be some level of trust, understanding, and awareness about the problem, to begin with.
In 2017, a non-profit associated with Bill and Milinda Gates, called Global Good, collaborated with Mark Schiffman (an epidemiologist from National Cancer Institute) on a project that set out to study an image collection for cervical cancer. They used a particular machine learning technique to find features, similarities, and dissimilarities in tumors. They began with cervix images of over 9,000 women collected from more than seven decades in Costa Rica. They had also accumulated 18 years’ worth of more accurate reports on pre-cancer and cancer diagnostic cases. They found that the machine showed about 91% accuracy in detecting pre-cancer, cancer, and healthy tissues. The new measurement has given Schiffman reason to pursue a cheaper and less cumbersome screening test for cervical cancer. And perhaps even a cell-phone based camera device that can help with early detection.
Another Stanford University research collected internet images of skin lesions. They worked with dermatologists to correctly classify over 2,000 skin diseases. This exercise was carried out to as patients very rarely have all the data needed for diagnosis.
AI in the arena of precision medicine would have innumerable possibilities if we removed the scarcity of drugs available for all patients and their overall efficiency. The technology also offers scope for start-ups in the field of drug discovery.
Research at the University of Dundee in Scotland, with co-partners Sanofi, GSK, and Exscientia, has aimed to identify symbiotic combinations of cancer drugs to develop drugs targeted at them eventually. There are only a handful of such endeavors, and this kind of research is still in its infancy. Only Benevolent AI, a British company, has come up with a reliable result that has led a drug candidate to move on to phase II of a clinical trial.
Tracking down the Evolution of a Tumor
As Ukrainian-American geneticist said, “Nothing in biology makes sense without evolution.” Cancer cells can overcome unfavorable conditions and prevail. This characteristic makes it hard to tackle them with drugs in clinics. Contemporary geneticists are exploring clonal diversity, drug resistance, and causation within an evolutionary framework. IBM has been ambitious with its AI platform Watson for Genomics, but nothing conclusive has been found yet
If oncologists can track down the mechanisms of drug resistance, they can classify patients into stratified categories and create precision treatment protocols. The results for this can be promising.
There are a lot of unstructured electronic records in health companies and hospitals, which can only be tapped into for useful insights if the current data is structured. Regulations for the access and proper use of data need to comply with data privacy and security issues as well as clarifications from different stakeholders ito varying levels from raw algorithms to results. Mechanisms need to be validated before its incorporation in the health sector, but for this many obstacles still persist. There is a wide gap between clinicians and data science experts.
AI has been through a revolution over the past decade. Breakthroughs in AI for med-tech and cancer will reduce trial costs. If invested strategically, they can detect tumors, diagnose disease, and even generate treatment recommendations in real-time for patients based on responses.