Just last week Datafloq covered the arguments on the Good vs Bad debate about the potential of Artificial Intelligence (AI). The application of AI in cancer prevention, detection, and treatment is certainly something worth adding to the good column.
Humanity has tried for years to find a foolproof cure for cancer but has been stymied at every turn. For all of human history, we have been plagued by this group of diseases. Despite attempts to cure cancer, diagnose the disease earlier, or even the U.S.’s 197 declaration of a War on Cancer, this disease manages to catch people by surprise and devastate lives.
AI is making significant strides in the fields of medicine and healthcare. Recently, the Harvard Business Review ranked its top 10 promising applications of AI in the healthcare industry. Among the top five were robot-assisted surgery, virtual nursing assistants, administrative workflow, fraud detection, and dosage error reduction.
Though the clerical and administrative applications of AI are definitely exciting for healthcare cost reduction and streamlining, the possibility of cancer eradication is today’s talking point.
AI’s Role in Detection and Diagnosis
The speed and agility with which AI can sort through data provide a huge advantage for early cancer detection. Scientists at Harvard Medical School and MIT are training AI to examine mammograms to differentiate cancerous cells from regular ones, long before human eyes could.
For particularly aggressive forms of cancer like mesothelioma, which affects internal organs with disastrous consequences, early and accurate detection can sometimes be the difference between life and death.
By tailoring these AI to specialize in different kinds of cancer, the diagnosis can be even more successful. Specific programs have been developed to diagnose skin, colorectal, prostate, neck, and breast cancers. A particularly successful AI developed by a team of French, German and American scientists can now diagnose malignant moles and spots with 8% more accuracy than dermatologists.
The success of AI on its own is impressive and is an important technological stride as a whole. Even more promising are the joint capabilities of machine and human intelligence. In an interview with CNBC, Dr. Andy Beck of the Harvard Medical School noted this interesting application of AI.
The combination of human plus AI in this example reduced the expert’s error rate by 85 percent, Beck said.
The Treat(ment) of a Lifetime
Tailoring a cancer treatment plan to a specific patient based on their DNA and RNA seems like a futuristic dream for many oncology teams. At the Mount Sinai Icahn School of Medicine, a team of oncologists is making this dream a reality.
Mount Sinai’s Director of Translational Research in Myeloma, Samir Parekh, says the team has dubbed this process multi-omics. It’s called that because the team uses both DNA, which carries the genetic code, and RNA, which carries the instructions for DNA’s work.
Using RNA as well as DNA allows scientists to look for any inconsistencies in genetic coding and expression that might drive cancer. Of course, careful examination and analysis of an entire person’s genetic makeup isn’t something the human mind could process without a large margin of error and a decent amount of time.
Enter, computers. Parekh’s team built a supercomputer for this exact purpose and named it Dr Crusher. The system took data from myeloma treatment and success from other patients and ran it against a database of DNA and RNA patterns to recognize any drug sensitivities or previous patterns given to a high likelihood of success.
In the end, Parekh told Forbes that 76% of their patients responded positively to Dr Crusher’s plans. Dr C isn’t the only machine making huge strides in treatment plan implementation. IBM’s Watson is also pushing boundaries, though the two programs approach treatment very differently.
Crusher examines genetic information and then scans databases for indicators of success or failure, while Watson uses medical history and records to learn and extrapolate how different treatments might affect unique patients.
Other systems, like Google’s DeepMind Health AI, have been trained to analyze retinal scans and search for abnormalities that would indicate a dangerous anomaly. The more scans fed to these systems, the more accurate their diagnoses become.
The Implications of Education
Continuing to use AI will increase the success of the machine’s work. The more data an AI is exposed to, the more it learns and the better it can successfully perform its task. This phenomenon – called deep learning – explodes the application potential of AI from the present into the future.
During treatment, AIs can analyze more factors related to a patient’s health background than even the best doctors. A successful AI could theoretically take into account genetics, family history and lifestyle to craft a treatment plan with the best chance of success before your physician has even asked if your family has a history of high blood pressure. Combining these potentials into one presents a huge possibility for advancement.
The broad application of this system, say over an entire hospital system, would allow AI to learn from other patients’ history as well as your own, and then extrapolate using the bigger sample sizes.
This kind of unsupervised learning, where a machine progresses from its own internalized information, and not when told to by humans, is something humanity will need to come to terms with. It represents the possibility for advancement across all sectors of human life, but also the likelihood of a shocking new world of regulation, monetization, and sophistication.
For now, the chance for faster diagnosis, lower rates of misdiagnosis, and personalized treatment plans are enough to champion AI and machine learning as the MVP of healthcare for the next few years.
