Dammit Jim, I’m a Doctor, not an AI – Healthcare and Machine Learning

PBlog09042018Healthcare costs continue to skyrocket. In 2016, healthcare costs in the US were estimated to be almost 18 percent of GDP. The healthcare industry is seeing an unprecedented and accelerating growth of ESI. This avalanche of data is being generated from the digitization of healthcare information, EHR systems, precision and personalized medicine, health information exchanges, new imaging technologies (DICOM), new regulations, IoMT (Internet of Medical Things), and other major technology developments.


AI (Artificial Intelligence) and ML (Machine Learning) are powerful tools that empower healthcare organizations to process the avalanche of data in near real-time, maximize the value of this data, and deliver deep insights, anomaly detection, patient experience enhancement, assisting in diagnoses, reducing medical staff workloads, and reducing costs. For example, a possible use case involves patient medical risk analysis based on accumulated data. What if an AI system could analyze a patient’s records before or after a visit and, based on physicians’ notes, recent blood test results, and email conversations between the doctor and patient, alert the doctor that the patient may be showing signs an urgent medical condition giving the doctor time to schedule new tests immediately?

On the payer side, healthcare insurance companies could use AI with patient history and current claims for in-depth analysis using predictive modeling technologies to improve, manage, and analyze claims data, increase claims performance, and determine possible fraud.


The fact is that a next-generation cloud infrastructure, like the Azure Cloud, can utilize cloud economies of scale to enable even small healthcare organizations to take advantage of emerging AI technologies at a much lower cost instead of attempting to do it themselves on premise. In reality, healthcare organizations are already utilizing cloud-based AI/ML to reduce costs, incorporate new AI/ML capabilities faster, and improve security, agility, and scalability.

Cloud technology provides healthcare organizations with advanced tools and safeguards so they can incorporate new technology to ensure patient privacy and security, and meet complex regulatory requirements and data protection laws.


The healthcare industry, both providers and payers, have suffered from fraud, waste, and abuse for many years. In fact, fraud represents a substantial portion of total annual spending on healthcare including overbilling, double and triple billing, bogus prescriptions, unnecessary and repeated medical procedures, unnecessary lab tests, and fictitious claims. What if a system could be observing healthcare processes and related activity and recognize and alert administrators and medical personnel about the possible fraud activity?


Using machine learning algorithms, the AI/ML application is trained using task profiles along with examples of related and unrelated data.

Supervised machine learning is the process of a computer algorithm learning from a training dataset with the interaction of human trainers. In the training process, the ML technician starts off with a data set and a known correct conclusion.  The algorithm iteratively makes predictions on the training data and is corrected by the teacher. This training/correction process can last through tens or even hundreds of training cycles. Learning stops when the algorithm achieves an acceptable level of accuracy and therefore ready to go live. Using this process, the AI is taught to make the best statistically relevant guidance.

In a supervised machine learning system, training cycles are extremely important in achieving results with the highest possible accuracy.

The high-level training process used by supervised AI/ML is as follows:

  1. Connect the AI with the healthcare data in the cloud – usually in the same system
  2. Provide example correct and incorrect content for use in the training cycles
  3. Begin the first training cycle by providing the system with related and unrelated data
  4. Once the AI provides an answer, provide feedback to the system on the correctness of the answer
  5. Provide additional content and begin the process again
  6. Once the system reaches the desired accuracy, it is ready to go live
  7. Healthcare personnel continue to interact with the AI to continue fine-tuning its accuracy

The number of training cycles could span from tens to hundreds of human-managed training cycles – a rather involved process. However, accuracy can be measured and proven– important when dealing with patient data.


The goal of unsupervised learning is to provide the AI with huge data sets and let the AI learn on its own, unlike supervised learning, there are no correct answers to train with, and there is no human teacher. The algorithms are left to their own devices to determine and present the correct conclusions.

However, supervised machine learning is the technology most used today due to its proven record of achieving extremely high accuracy rates. But, unsupervised machine learning is making of headway and will eventually replace supervised learning systems.

To read the complete blog, click here


Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s