A.I. in Medicine
The implementation of A.I. into common medical procedures won’t eclipse the need for qualified doctors; rather, they work better in conjunction, correcting each other’s errors.
Reading Time: 3 minutes
In recent years, artificial intelligence (A.I.) has proved to be a sensitive topic, especially when considering the ethical dilemmas surrounding its medical applications. Nonetheless, brain surgeons are bringing A.I. and new imaging techniques into the operating room to diagnose tumors. Recent efforts to implement this state-of-the-art technology have focused on analyzing tissue samples of patients undergoing surgery in order to guide further treatment. A major concern with the traditional method of sending a biopsy to a lab for analysis is efficiency: it takes more than 20 minutes while the patient is under anesthesia. As prolonged sedation is undesirable, reducing the time needed to analyze the biopsy is a top priority. The new method, taking at most three minutes, is extremely promising.
This new technology called stimulated Raman histology employs lasers to scan tissue samples with different wavelengths of light that each scatter uniquely for different types of tissue. When the light hits a detector, the detector ascertains the light’s scatter pattern and emits a signal that a computer can use to reconstruct the image and identify the tissue. This enables neurosurgeons to, in the case of brain cancer, distinguish normal from cancerous brain tissue. Also, unlike the traditional method, this technology does not destroy the tissue sample, allowing it to be used again for further testing. This technique can detect details as small as a tumor spreading along a nerve that the traditional method may miss, leading to fewer false positives and negatives.
A clinical study funded by the National Cancer Institute and published earlier this year involved brain tissue samples from 278 patients that were analyzed by both deep neural network algorithms and neuropathologists during surgery. The result was a draw: the pathologists were correct 93.9 percent of the time and the algorithm was correct 94.6 percent of the time. Such misdiagnosis is a problem that plagues medicine as a whole, not just neurosurgery. For example, traditional screening mammograms miss about 20 percent of breast cancers, including false positives and negatives. Clearly, just as brain surgeons need a better method for analyzing tissue, oncologists need a more reliable way of analyzing mammograms.
Luckily, similar steps have been taken in the interpretation of mammograms for breast cancer by researchers at Google. A machine learning algorithm was trained on a dataset of 100,000 images whose diagnoses were already known. The system did better than radiologists, having lower percentages of false positives and negatives by as much as nine percent. Interestingly, both of these systems made mistakes that the pathologists didn’t and vice versa. These systems won’t be interpreting scans and biopsies on their own anytime soon; however, their accuracy suggests that rather than replacing doctors, A.I. can assist their judgment by participating in doctors’ routine practice in reading these scans. The combination of an algorithm and human intuition can only improve the ability to make correct diagnoses; each can catch the other’s mistakes. After all, robots don’t get tired or face clouded judgment after a long day of work.
Nevertheless, putting robots in charge of many patients’ healthcare comes with its flaws. Dr. Eric Topol’s book “Deep Medicine” weighs the benefits and drawbacks of A.I.’s recent intrusions into medicine. A notable con is that the highly technical nature of the systems’ algorithms makes explaining medical procedures to the patients’ families hard, inevitably making them uneasy about their loved ones’ healthcare. Dr. Topol is also adamant that robots can never replace surgeons, alluding instead to a sort of competitive exclusion in which they take on a new role. He says that they can provide “human expert contextualization,” interacting with patients and acting as the mediator between life and death that machines lack the humanity to be. In fact, he thinks that A.I. can save doctors from increasing levels of burnout and depression from doing tedious tasks like reading scans and taking notes. Productivity will skyrocket with doctors being less pressed for time when connecting with patients.
Recent advances in A.I. are making significant headway in medicine. Deep neural networks are cutting down time needed for crucial brain tissue analyses, and machine learning algorithms are analyzing mammograms just as well as professionals. However, judging by their limited accuracy and the clear ethical barriers inherent with putting robots at the helm of healthcare, A.I. in medicine still has a ways to go before being accepted as common practice.
Links
https://www.nytimes.com/2020/01/06/health/artificial-intelligence-brain-cancer.html