The rise in data, processing power and algorithms have made AI, or more commonly machine learning, useful for more that theory. Electronic Health Record (EHR) adoption in the US is over 80%, while Europe is catching up. This adoption, in combination with increased standardization, has made it easier to take advantage of the gathered data. We can speed up the time it takes to develop a new drug, help diagnose patients, and even keep recovering addicts avoid overdosing.
The cost of developing a new drug from start to finish is over a billion dollars, and on average, takes over ten years. This process includes many stages, where testing determines whether or not a particular solution advances to the next stage or is done away with. Automating some of these tests decreases development times. Savings lie in a few specific solutions. Firstly, we can hypothesize more potential compounds, compounds that have a higher likelihood of working. Furthermore, tests can be automated and anomalies detected. We can discover particular patterns when looking at sample groups and help fight biases in our research. Drug discovery will still be a difficult and expensive process, but technological advancement continues to provide more efficient solutions.
Experiments and new projects are not the results of an individual effort, but of a team effort. And teams need to share information with each other. Data, experiments, and algorithms are not interpreted in an intuitive manner. Visualizations are the best communicative effect, allowing team members to share their findings and status more adequately. Furthermore, dashboards allow you to manipulate inputs, zoom into particular graphs or charts and have greater control leading to deeper insights. At Appsilon, we have put in a lot of effort into open source Shiny packages for better UI and UX within Shiny dashboards. We have also scaled Shiny applications without Shiny Server to hundreds of users. If you or your team are sharing results and have the nagging feeling that there has to be a better way, take a look at our technology page, where you will find a few of our Shiny packages.
Medical imaging is one of the most useful realms of healthcare. It provides a noninvasive view into a patient and has become a necessity for diagnosis. Medical imaging includes CT scan, X-rays, ultrasound, PET and more. These methods rely on a manual review. Doctors, nurses, and technicians are aware of what to look for in these scans. Their time is valuable and can be better spent elsewhere. Deep learning has achieved better than human levels of image recognition.
These models and algorithms can be further applied to medical imaging for classification, anomaly detection, and segmentation. We are not looking to replace doctors, but support them in their work and increase the quality of patient care provided, while at the same time, keep costs down.
Patient notes, prescription orders or diaries are a few of the many types of unstructured and often handwritten or hand typed data points in healthcare. You may be surprised to find out that handwriting recognition has been used since the late 80’s. Yann Lecun developed a solution for zip code detection. We have come a long way since then, allowing to us to recognize even a doctor’s handwriting. Furthermore, we can make sense of the text that has been entered. Patient files can become much more insightful as we begin combing structured with unstructured data. Below, is a diagram of data points that can be associated with a patient’s health records and potentially helpful in diagnosis and treatment.
We are all animals, creatures of habit. Addicts are aware of their predisposition to relapse all too well. The most common cause of an overdose is when a recovering addict relapses and does not account for their decreased tolerance. With such a high potential for loss of life, we must do all in our power to help those who make mistakes and unwillingly approach death. But solace lies in our habits. Certain behaviors or patterns are indicative of higher risk behavior. With the help of machine learning, we can model historical data and behavior to help us predict and identify which patients need more support with greater accuracy than results based solely on human intuition.
The influx of data points means that doctors are more informed about their patients than ever before. A general practitioner, for example, can prepare for his day by going through the records of the day’s patients. This can be used in telemedicine, a growing aspect of care. Patients can also be diagnosed before even arriving at the hospital or office. A few converging questions can give a range of what we can expect and which further tests may be needed. Furthermore, the diagnostic predictions we receive can become more precise over time, as doctors feedback can be used to further train and fine tune the algorithm.
Healthcare is one of the most data intensive realms of our society. Wherever we find data, even in the form of a medical notepad, we can apply data science to improve accuracy and alleviate some of the repetitive workloads. This list is not comprehensive, as medical applications of machine learning in healthcare even include cancer detection.
Get in touch with us to discuss how data science can help you and your patients.