Machine learning in Healthcare

Clinical bioinformatics is a new emerging science combining clinical informatics, bioinformatics, medical informatics, information technology, mathematics, and omics science together. Because it’s easy to scale up computer as compared to doctors and computers can run 24/7.

Machine learning in Bioinformatics

Machine learning is a process used to teach a machine to work accordingly. First a program is written using math and a formula is created, which computer can understand. This formula distribute scores depending on the given condition, say for example we are training a robot to walk, this formula distribute numbers like +1 for correct step, -2 for unstable move and so on. Then depending on program computer can learn either by experimenting itself and moving forward in the direction which maximize the score, or it can learn by seeing others. In case of heath care bioinformatics scientists uses the second method i.e. training using available medical records.

Here is an example that will help us get a better grasp of what I am explaining.

TREWS (Targeted Realtime Early Warning System)

TREWS is developed by Dr Suchi Saria, she is a machine learning and healthcare professor at Johns Hopkins University. Dr Saria with his colleagues develops a software known as TREWS. TREWS is a machine learning program which is designed to detect the sepsis with high accuracy and efficiency.

Sepsis: (body releases chemical to fight infection which generates negative inflammatory response leads to organ failure). The body normally releases chemicals into the bloodstream to fight an infection. Sepsis occurs when the body’s response to these chemicals is out of balance, triggering changes that can damage multiple organ systems.

The problem with sepsis is shown in a study conducted by Harvard, it shows when experts are shown cases with or without sepsis their predictions are not in sync.

Case study

Mrs Mani a 52 yr old women. She is admitted in hospital with footsore, because no medical complication is in sight so doctors place her in general ward. After 3 days theirs a symptom of pneumonia due to which doctors recommend her antibiotics. On day 6 tachycardia is detected and on day 7 she get sepsis shock and transferred to ICU. Slowly her organs stared failing, and on day 22 she died. She may have survived if diagnosed earlier.

When TREWS run on Mrs Mani’s medical records it identify sepsis 12 hour prior to the shock, which is a significant amount of time (every hour chances of death increase by 7–8%).

TREWS is a ML program which learn by studying electronic health record with and without sepsis. Example Creatinine is a waste which is filtered out by kidney, but in case of sepsis kidney’s ability to filter it out is decreased. But there may be other causes also, that’s the task of TREWS to find out weather the level of creatinine in blood is increased because of sepsis or other reason.

It is tested on 16000 patient data, on an average in most cases it detect 24hr prior to shock. 2/3 patient prior to any organ dysfunction. 60% increase in detect performance.

TBM (Transport Based Morphometry)

Imaging (X-ray , MRI, ultrasound etc) is the fastest growing medical data. Imaging sometimes reveals a disease before one can feel it. In disease like cancer the earlier you diagnose the more likely person survives. It is important to diagnose disease in early stages, it can save lives, cost and suffering.

The problem is when we turn back the clock the visible evidence become smaller and smaller and finally invisible to human eye. Small changes that eye cannot see still exists. These small changes can be decoded with the help of computer.

Osteoarthritis is one these diseases which can detected way back even before the pain started. Osteoarthritis occurs when the protective cartilage that cushions the ends of your bones wears down over time and cause bones to grind together, causes heavy pain. About 10% of 60+ age population is suffering from osteoarthritis.

Presently doctors can see osteoarthritis only after damage and severe pain. But in this study it is shown that this disease began long before.

Comparing MRI of knee cartilage even experts can’t tell anything about osteoarthritis, because every person knee have a different structure and human eye get saturated with these individual changes.

Dr. Shinjini kundu with her team develops TBM a machine learning program which learn using available MRI of people who develop osteoarthritis down the line and who don’t. TBM observes every single pixel in MRI and capture changes which are invisible to human eye. TBM has 86% accuracy rate in identifying samples who may develop osteoarthritis 3yr down the line.