There’s a lot of talk around the recent rise of machine learning (ML) and the promise it holds for oncology. Machine learning is quickly becoming a vital component of every-day medical testing, almost like a regular blood test, an MRI, or the old-fashioned stethoscope. ML assists healthcare providers, physicians, nurses, imaging technicians, and others in improving patient diagnosis in areas previously deemed as undiagnosable “dead-zones.”
Machine learning represents the most successful branch of artificial intelligence–development of programs with the capacity to learn from data, playing a prominent role in facilitating novel applications that rely on the identification, correlation, and classification of complex data patterns for patient stratification.
In the past, clinical testing and analysis were done manually, a mundane and tiring job for a human being, often resulting in important findings being missed due to exhaustion. Machine learning is a tool that never gets tired. It can assess thousands of images and realms of data with sustained quality and reliability, enabling clinicians to see things—in imaging, blood tests, and big data—that otherwise might have been unseen.
What’s the biggest risk with machine learning? If a mistake is made while teaching the machine, and no validation occurs, the results can be completely inaccurate, resulting in loss of prediction performance. The challenge with machine learning is ensuring that when developing products, the algorithm is tested multiple times, and capabilities are constantly being validated. Sometimes this involves years of training and retraining, but the benefits for improved diagnosis far outweigh any potential disadvantage.
Machine learning in precision oncology
Precision medicine holds immense promise as an enabler of targeted therapies and a healthier society. Machine learning tools are driving various areas of medicine, but the scope and power of ML in oncology is substantial and is facilitating the implementation of further personalized precision medicine approaches. Over the past few years, ML’s potential in precision oncology has become more apparent by the reporting of major advances in deep learning (DL), and its application to a variety of diagnostic, prognostic, and other predictive tasks.
Machine learning capabilities offer clinicians the opportunity to more carefully tailor early cancer interventions—whether treatment-focused or preventative in nature—to each individual patient. Taking advantage of high-performance computer capabilities, machine learning algorithms can now achieve reasonable success in predicting risk in certain cancers by assessing multidimensional clinical and biological data. It achieves this in a few ways:
The emergence of machine learning in medical imaging has allowed for the greatest disruptive technology in decades. Not only has it given clinicians the ability to identify cancerous tumors at an earlier stage, but it can also detect and classify lesions, automate image segmentation, analyze data, reconstruct images, and more.
Precision medicine also has customized healthcare for cancer patients by stratifying patients into those who may respond to a treatment and those who may not by using diagnostic and prognostic biomarkers. Through analyzing biomarkers, ML approaches are enabling us to obtain explanations for patient-specific predictions. This provides clinicians a general view of which prediction features are important in order to assign a patient to a specific clinical outcome.
Proteomic profiling is a relatively unexplored area of machine learning. The aim of proteomic profiling in oncology to identify biomarkers and to aid in diagnosis, prognosis, and treatment of cancer by analyzing proteins. Due to the vast quantities of proteins needed to be assessed at one time, including cross-reactivity and huge variability among the different proteins, the challenge of proteomic profiling is immense. But proteins, in contrast to DNA and RNA, are the actual drivers of the biological processes in our body. If it’s there, it has a role, and this is where machine learning comes in. ML algorithms are being developed and trained on how to relate to specific proteins and assess them for signs of malignancy in the body.
In addition, machine learning capabilities can identify potential drug targets, advancing the development of novel therapeutic strategies.
One of the major obstacles in clinical oncology is resistance to therapy. In recent years, studies have indicated the contribution of the tumor micro-environment, as well the response of the body to anti-cancer treatment, as a key player in therapy resistance.
Using advanced machine learning and bioinformatics technology, together with a high-throughput protein analysis program, host (patient) response to treatment can be characterized, analyzed and predicted, enabling personalized treatment strategies with improved results and reduced side effects.
So, in machine learning we trust?
Machine learning offers opportunities to advance and accelerate precision oncology, giving researchers the opportunity to not only face new biological questions, but to also improve the techniques used at different steps of the profiling process. Without applying machine learning, achieving the full potential of precision medicine is impossible.
Although machine learning in oncology is only at its genesis, the future is bright. Researchers, healthcare organizations, and companies are all developing ML-based tools for various applications with the common goal of improving clinical outcomes for patients with early detection, more accurate diagnosis, and personalized treatment strategies. Implementation of ML tools in clinical settings is already happening, with imaging interpretation algorithms identifying lesions earlier and more accurately than humans ever have. Over the next few years we’ll see additional development and implementation of biomarkers, proteomic profiling and host response analysis.
The road to complete trust is still long, but the combination of the computational power of machine learning, together with the empathy and non-linear decision making of humans, carries tremendous promise. The global healthcare community look forward to seeing machine learning further precision oncology research, allowing us to develop personalized strategies to maximize the success of cancer therapy for every patient.
Ofer Sharon, MD, is a physician and entrepreneur with more than 17 years of experience in the oncology drug development industry. He is the CEO of OncoHost, the global leader in host response profiling for improved personalized cancer therapy.