The use of artificial intelligence technologies in medical diagnosis has been around since the emergence of machine learning in 1950s (Symbolic learning, statistical/pattern recognition, and neural networks). However, it is only with the maturation of pattern recognition branch of machine learning that considerably accurate results were obtained for medical diagnosis. And while the field of machine learning in medical diagnosis continue to be heavily researched, a few challenges will continue to depress wide spread application of these technologies.
One of the major challenges in the application of machine learning in medical diagnosis is the large variety of documented medical conditions. Currently, there are over 15,000 documented pathologies/conditions/diseases/infections, and each one has a nearly infinite number of variables. No amount of data is going to grant a machine the intuition required to know how to proceed with such an amount of data and variables. Secondly, diseases sometimes manifest themselves in almost similar symptoms, this can be entirely confusing to machines (Read the Chinese Room argument).
The use of machine learning in pockets of medical field have however shown remarkable progress. Currently there are a number of application of machine learning such cancer diagnosis, tuberculosis diagnosis, etc. These, albeit, have shown remarkable promise due to the limited focus of such applications.
According to Igor Kononenko (Machine Learning for Medical Diagnosis), specific requirements for machine learning systems are:
Good performance: The algorithm has to be able to extract signi¯cant information from the available data. The diagnostic accuracy on new cases has to be as high as possible.
Deal with missing data: In medical diagnosis very often the description of patients in patient records lacks certain data. ML algorithms have to be able to appropriately deal with such incomplete descriptions of patients.
Dealing with noisy data: Medical data typically suffer from uncertainty and errors. Therefore machine learning algorithms appropriate for medical applications have to have effective means for handling noisy data.
Transparency of diagnostic knowledge: The generated knowledge and the explanation of decisions should be transparent to the physician. She should be able to analyse and understand the generated knowledge.
Some of the most used ML algorithms include Bayes classification, K-means, K-nearest Neigbour.
We find that with the loads of data that is available for diagnosis of malaria, especially in Kenya, we could train an algorithm to be able diagnose Malaria from user symptoms. We chose malaria because:
Most of these deaths from malaria are caused by not sleeping under treated mosquito nets, increasing resistance of malaria parasite to anti-malarial drugs, and the inability to access treatments within 24hours from the onset of symptoms. It is the later that we aim to help resolve. We think that by self diagnosis, people may well be able to visit nearest health facility sooner.
Zahanati app, developed by Doban Africa, is a research project that aims to reduce incidence of late diagnosis of malaria by availing tools for self diagnosis. This is not meant to replace doctors or medical officers or usurp their roles but to provide information only. The app uses artificial intelligence to determine incidences of malaria. The app is currently running on mock data from medical student diagnosis of malaria and is in beta testing and will be available in all major platforms. The beta release runs on android as JavaMe platforms.The data collected from these endevours (which is de-identified, as we don’t collect any personal information) will be opensource and can be used to research and map incidents of malaria and other symptoms across the country.
You can download the apk file below.
If you have questions or would like to participate in the testing or research, please drop as mail at firstname.lastname@example.org with your specific details of how you would like to participate.