OR WAIT null SECS
3 cases that highlight how Python & machine learning could optimize medicine.
It’s no secret that open source is driving significant machine learning (ML) innovation. At the heart of this trend is Python, which is widely considered the tool of choice for data science projects in general and ML initiatives in particular. This piece will look at the use of Python-based ML in healthcare in three specific areas.
ML has been a component of healthcare research since the 1970s, when it was first applied to tailoring antibiotic dosages for patients with infections. But with the increased volume of electronic health records (EHRs) and the explosion in genetic sequencing data, healthcare’s interest in ML is now at an all-time high.
According to McKinsey Research, big data and machine learning in pharma and medicine could generate a value of up to $100 billion annually. That’s based on better decision-making, optimized innovation, improved efficiency of research and clinical trials and the creation of new tools for physicians, consumers, insurers and regulators.
How does Python fit into this picture? It’s the go-to language for many developers, ranking as one of the most popular programming languages and used widely across various tech disciplines, from data engineers to web programmers. Python’s rising popularity also touches data science and ML.
And the emergence of open-source language automation presents tremendous opportunities in healthcare for Python-based ML. Python language builds can be completed in minutes with specific ML packages and be vetted for open-source security and licenses. What’s more, Python now features the bulk of all open-source ML and data engineering tools. Developers can use the language to efficiently build innovative solutions while ensuring that code is secure throughout the life cycle of the applications.
Hospitals and clinics are strongly resource-constrained, making cost control critical to sustainability. And ensuring medical staff, treatment and diagnostic facilities are scheduled efficiently is a large-scale optimization problem with many dimensions. Doctors need to identify patients who are not following their treatment protocol. Patients undergoing surgery need skilled staff to care for them, sometimes around the clock.
ML can play a role in all of this, from predictive inventory management to improved triage for emergency departments to patient surgery and care. That’s why industry analysts at Accenture estimate that by 2026, the ML health market could potentially save the U.S. healthcare economy $150 billion in annual savings.
Diagnostic errors have been linked to as much as 10% of all patient deaths and may also account for between 6% and 17% of all hospital complications. ML is one potential solution to diagnostic challenges, particularly when applied to image recognition in oncology (e.g., cancer tests) and pathology (e.g., bodily fluid tests). In addition, ML has also been shown to provide diagnostic insights when examining EHRs.
When it comes to correctly analyzing medical images, ML success rates of up to 92% sit just below senior clinician success rates of 96%. However, when ML diagnoses are vetted by pathologists, a 99.5% accuracy rate is achieved. And even more promising is the use of ML to provide diagnoses based on multiple images, such as computerized tomography (CT), magnetic resonance imaging (MRI) and diffusion tensor imaging (DTI) scans. The human brain has a hard time integrating these different views into a whole, but ML solutions were better be able to process each unique piece of information into a single diagnostic outcome.
Even the best medical practitioners struggle with predicting how diseases will progress. In fact, it’s more guesswork than science, especially when it comes to terminal illness. Existing solutions help improve patient treatment by better predicting disease prognosis. But these solutions are either too costly or too time-consuming to implement in a practical manner. What’s needed is a solution that provides better predictions more cheaply and quickly than existing methods.
With recent breakthroughs in artificial intelligence (AI), predictive prognosis solutions have turned to Python-based ML techniques for an answer. This type of solution has been used to predict the mortality of a patient within 12 months of a given date based on their existing EHR data.
Python was used to create a deep neural network (DNN) using Pytorch and Scikit-Learn in order to predict death dates for patients with terminal illnesses. Each patient’s EHR was put into the DNN, including current diagnosis, medical procedures and prescriptions. The DNN then provided results that allow doctors to bring in palliative care teams in a timelier manner.
As machine learning creates disruptive change across many industries, it only makes sense that it would be applied to the healthcare industry. With billions of dollars to be saved and better care to be delivered, the field is increasingly turning to ML. They are doing so most often via Python, the open-source language that many consider the best suited for ML initiatives. Developers can build solutions that benefit human health and well-being, confident in the knowledge that their work will be secure.
To optimize Python for ML in Healthcare, considering how to use, monitor and secure the language code or implement open-source language automation will be of benefit to companies to manage the risk and accelerate the innovation Python can deliver.
About Bart Copeland:
Bart Copeland is the CEO and president of ActiveState, which is reinventing Build Engineering with an enterprise platform that lets developers build, certify and resolve any open source language for any platform and any environment. ActiveState helps enterprises scale securely with open source languages and gives developers the kinds of tools they love to use.
Get the best insights in digital health directly to your inbox.