- We have combined traditional machine learning with so called deep learning, says Michael Alexander Riegler.
He is a scientist at Simula Research Laboratory and recently got his Ph.D degree at the Department of Informatics on a multi media system capable of making better diagnoses.
You can also read this article in Norwegian
Quicker diagnosis
- The accuracy is at least as good as in today’s systems, but results can be obtained much faster. That means the diagnoses can be made in real-time, sier Riegler.
The system is called EIR, from the Norse goddess who ruled over medicine.
So far Riegler and his colleagues have gathered data from the eight most common gastric illnesses and anatomical landmarks, meaning colon cancer, gastric ulcer, celiac disease, inflammations (Crohn’s disease and ulcerative colitis) and chronic diseases, and gastro-oesophageal reflux disease - where stomach contents come back up into the esophagus.
All these diseases have great impact on the patients’ quality of life.
Rapidly growing form of cancer
Riegler points out that three of the six most common cancer types occur in the gastrointestinal system. We are talking about 2.8 million new occurrences world wide each year, with a death-rate of 65 percent.
- Colon cancer is the most common form of cancer for Norwegian men, and the second most common for Norwegian women. This cancer type is the most rapidly growing in prosperous countries. An early diagnosis is crucial. Rapid treatment may mean the difference between life and death.
The health data needed to identify the different diseases come from several partnerships. Riegler and the other researchers cooperate closely with several health institutions - primarily the Cancer Registry of Norway, Bærum Hospital and Karolinska University Hospital.
- We have shown that EIR can conquer «state-of-the-art» systems both in the automatic analysis and when it comes to the processing speed, says Riegler.
Ever-increasing accuracy
EIR takes the accuracy level to 93 percent, with a speed of 300 pictures per second.
- With even better data sets fed into EIR, we can reach an accuracy of 95-99 percent in the future, Riegler estimates.
These promising results have led to contact from more hospitals wishing to cooperate with Riegler and the rest of the team. Partnerships have been made in Italy, the US, Spain, Japan and Sweden in addition to Norway.
- We have had so many applications from institutions wanting to cooperate that we had to turn some down for capacity reasons, says Riegler.
Machine learning and open source code
He finds it very powerful to combine different technologies within artificial intelligence and machine learning.
EIR is based on a combination of search based classification, which has to do with methods for re-tracking information (like Google image sarch) - and machine learning.
The system is alsomade open source, meaning that anyone can use it and will not be locked out by suppliers’ proprietary solutions.
- Doctors will not be superfluous
Riegler points out that some doctors perhaps fear being superfluous, like other occupational groups that risk being outnumbered by artificial intelligence.
On the contrary, Riegler thinks that it will be impossible to substitute doctors with automatic systems, but these systems will help the doctors.
- EIR will make it easier for doctors to quickly make the diagnosis, so that they can care better for the patients and have more time for other patients. We also know that people who are diagnosed need professionals to talk this through with.
Frequent examinations
He reckons a lot will happen in the diagnosis and treatment of gastric illnesses.
An annual check of the gastrointestinal system for people over 50 years has been discussed. Riegler means this can be done by swallowing a pill with a camera, which will then take a ride through your system.
If any suspicious pictures are shown, the person will be summoned for further examinations.
- Maybe this can be a reality within ten years, the Simula researcher hopes.
More information:
Riegler's PhD thesis: EIR – A Medical Multimedia System ffor Efficient Computer Aided Diagnosis
