While doctors, nurses and medical officers will probably never be replaced by robots and computers, modern technology, machine learning, and AI are already transforming the healthcare industry. This transformation is leading to improved outcomes and changing the way doctors think about providing care. A recent study by industry analysts at IDC predicts that 30 percent of providers will use cognitive analytics with patient data by 2018. At the same time, global spending on cognitive systems will grow from nearly $8 billion in 2016 to more than $47 billion in 2020. We’re talking about significant numbers here!
But what exactly is the machine learning process?
It is the process that involves computer algorithms that are able to learn highly complex and intricate relationships in a multi-dimensional ocean of data. The benefits of the machine learning process can be wide-ranging as it provides:
- Massive data input from unlimited sources – it can consume, review and adjust to virtually unlimited amounts of detailed data
- Rapid processing, analysis, and predictions – it can support the ability to act, optimize and modify processes in real time
- Action systems – it can create models and experiences that are much more dynamic and customized
- Learning from past behaviors – it can adjust its approach based on experience, thus offering a much more personalized solution
Machine learning and predictive analysis in healthcare
Advancements in medical technology have dramatically increased the quantity of data available in the healthcare industry, ranging from patient reports and genomic data to electronic medical records and more. These data can provide a wide range of insights into patient cases for prevention and cure. Predictive analytics leverage these large, heterogeneous data sets to further knowledge and foster discovery. It can identify similarities among patients based on a wider and deeper range of variables or characteristics (age, genetic marks, previous illness, etc.), leading to finer groupings and thus more accurate predictive analysis.
Such predictive modeling can facilitate appropriate and timely care by forecasting an individual’s health risk, clinical course, or outcome. Uncovering these types of patterns at an early stage can be of great benefit to both doctors and patients, and can explain the use of drug combinations and treatments independent of any promotion by mapping the predicted targets on the disease signaling network. Compared to statistical methods, machine learning can increase prediction accuracy (sometimes doubling it) with less strict assumptions, for example on data distribution.
Digging deeper, machine learning methods have already been employed in breast cancer screening in order to identify malignant and benign microcalcifications for predicting breast cancer survival and to model prognoses of breast cancer relapse. Given a large number of patients being treated for breast cancer, and the fact that most of the adverse effects of chemotherapy are cumulative, identifying patients who are not responding to a particular treatment allows for switching to a potentially more effective regimen and avoiding unnecessary side effects.
Recent studies show that predictive modeling using readily available clinical and quantitative MRI data shows promise in distinguishing breast cancer responders from non-responders after the first cycle of NAC (neoadjuvant chemotherapy). In this research, twenty-eight patients with stage II/III breast cancer were enrolled in the study and completed at least two of the three scans to provide usable data for the analysis. Eleven clinical variables available before initiation of the first cycle of NAC were used for generating the predictive models.
Figure 1. List of pretreatment clinical variables with a short description
As a result of the studies, three datasets were made and used for creating prediction modeling. The final conclusions of the study showed that predictive modeling approaches based on machine learning, and using the available clinical and quantitative MRI data scan, can be a reliable source for distinguishing breast cancer responders from non-responders after the first cycle of NAC.
Similar ML algorithms were used in research to predict two-year leukemia-free survival in cord blood transplantation – Random Survival Forest (RSF).
RSF is known to be adaptive to data, as it automatically recovers nonlinear effects and complex interactions among variables, and yields nonparametric predictions over test data. Variables in the study were also ranked according to their predictive contribution, as disease status, age, and TNC (total nucleated cell dose) count were found to be the most important factors.
Figure 2. A general schema template for predictive model building and evaluation
In general, most studies until now showed a predictive accuracy of machine-learning processes of about 75 percent – not high enough for widespread clinical use. This month, however, a new approach to imaging analytics driven by machine learning algorithms reported rates that exceed other automated methods and even rival human pathologists – the machine learning tool was able to achieve 89 percent accuracy, compared to just 73 percent accuracy from a human pathologist who spent 30 hours poring over 130 slides.
Researchers still have a lot of work to do in order to develop reliable, accurate, and highly sensitive algorithms that can become meaningfully integrated into the patient care process, but test cases and pilots are bringing vendors ever closer to routine clinical care. As these technologies develop, new and improved treatments and diagnoses will save more lives and cure more diseases. The future of medicine is based on data and analytics.