Machine Learning Example

Typical example of machine learning

Currently a patient comes to a doctor's office with certain conditions. The doctor using his/her knowledge, will diagnose and provide required medical prescription or recommend new testing or meeting a specialist.

Medical Visit

Currently doctor's office staff take basic symptom information from the patient and convey it to the doctor. Only generic information is collected by a automated system as a fixed set of menu options. In future, a patient will login to virtual doctor system designed with advanced AI and ML features. He/she will input medical conditions he/she is experiencing. As an example the patient inputs four conditions. As soon as all the conditions are input, the machine learning models start analyzing the input.

1. Model[1] - 4/4   ->  100% match (basic condition)
2. Model[2] - 4/5   ->   80% match (no prescription medicines required)
3. Model[3] - 4/10  ->   40% match (mild condition)
4. Model[4] - 4/50  ->    8% match (severe condition)
5. Model[5] - 4/100 ->    4% match (critical condition)
6. Model[6] - 4/500 ->  0.8% match (emergency condition - ER admission)

Patient-1: From the above model analysis, the data input is close to Model[1] and Model[2] based on the data match. Behind the scenes, the match percent will be computed based statistical, AI and ML algorithms. The system can make a decision about treatment. It can create a simple diagnostic report and recommend OTC medicines if required and also recommend subsequent visit if conditions change or get worse in specific time frame by data transmitted by automated continuous patient monitoring systems. It will also send notifications to doctor, health monitoring devices and patient.

1. Model[4] - 15/50  ->  30% match (critical condition)
2. Model[5] - 15/100 ->  15% match (emergency condition)
3. Model[6] - 15/500 ->   3% match (emergency condition - ER admission)

Patient-2: In above case there were 15 conditions entered by patient and matched by the virtual doctor system. The system started with Model[4] and higher. Although the match is estimated as 30%, the system is programmed to analyze using a related models to pin point the condition due to condition being rated as critical.

1. Model[4.01] - 15/20 -> 75% match (critical condition - specialist review)
2. Model[4.02] - 20/25 -> 80% match (critical condition - lab work required)

The system was able to match 15 in Model[4.01] and using statistical methods, it able to match 20 conditions in Model[4.02]. The system was able to recommend lab work (blood tests, biopsy, etc.). In real situations, the doctor will be notified along with other human interactions. But for minor ailments and conditions, AI/ML can perform well. The concept of waiting for days or months to get doctor's appointment will be simplified.

The models will be based on data science, statistical concepts, forecasting methods, time series, ARIMA, SARIMA, to name some well known formulations. More appropriate models and algorithms will be used in real scenarios. The data science, AI amd ML are constantly evolving.

If we visit Mayo Clinic, CDC, NIH and other medical research institutes, we can get a idea of number conditions that will be associated with a single ailment. Using those base conditions and machine learning concepts, deep learning, new and related models can be derived. In clinical trials AI/ML algorithms are used to analyze data and arrive at conclusions faster.