Research Paper

Machine Learning and Medical Diagnosis: Affordable Medical Care

It is estimated that there are currently over thirteen million children under the age of five that die from illnesses that could have been easily treated (“Right to Healthcare around the Globe). Many areas in the world, such as the Central African Republic, lack the resources or money. Machine learning used in conjunction with medical diagnosis could help these areas have a jump start in medical knowledge and resource. Creating a program that automatically diagnoses patients would decrease the amount of money needed to support areas, increase the medical technology, and create a better connection between engineers and medical professionals.

Machine Learning is a field of computer science that uses pattern recognition and computational learning theory. A program can be set to learn chess through watching/playing thousands of iterations of the game, and learning every minute detail of the games at a considerable speed. With machine learning, a computer can try and make predictions on what will happen based on what has happened before; it predicts the future by looking at the past trends. In other words, the computer will operate without being explicitly told what to do.

It is without a doubt that developed countries should invest finances into helping out those that are developing. In Central African Republic, CAR, there are currently many issues including “armed conflict, [limited] access to health care, and an epidemic” (MSF, 2013). With an ongoing conflict between armed groups and communities, an “effort to promote CAR’s national unity and mend its social fabric” has been obstructed (International Crisis Group, i). With the armed tension, “over 70 percent of health facilities have been damaged or destroyed, and there is a shortage of trained health care workers” (MSF, 2013). It is clear that there is a medical crisis that is not being amended at a fast enough speed.

Currently, in order to aid countries or areas, we need to spend a lot of resources to get doctors across the globe, in working order, with some sort of medical facility. For example, Doctors Without Borders confirms that they use more than 85 percent of their donations to fund program activities (Donation FAQ, 2013). Expenses include emergency and medical programs, program support and development, field staff, and communications (Donation FAQ, 2013). In 2010, the revenue of the group, “all raised through private contributors, such as individuals, foundations, and corporations” totaled to 264 million dollars (Donation FAQ, 2013). Simple calculation would give us that the Doctors Without Borders spends approximately 224 dollars in their programs. Using artificial intelligence and machine learning AI, it will be possible to create a means of medical diagnosis that is cheaper to fund and to implement in countries. Less money spent in one area means more money to spend to help out the rest of the developing nations.

This program will be targeted for users that are in areas that lack resources. These resources include funds to build any sort of medical facility, or areas that lack any sort of technologies to aid themselves with. This program will be designed so that it can be implemented at a low cost to provide affordable medical diagnosis to patients. The costs would include costs to manufacture the program, resources to implement these programs in countries such as CAR, and resources to help maintain the program. With the implementation, areas in need can use the program as a baseline for medical knowledge and a foundation to build a reasonable medical facility. This program can also be implemented in countries that already have a stable medical foundation. It can be used by casual patients who don’t feel the need to physically visit the doctor, but are somewhat worried about their health.

The program will be operated based on two main users: the patient and an administrator. The patient enters information about them self, and receive a diagnosis. The administrator would be in charge of defining new diseases, finding an appropriate database of training sets, and training the system (Kampouraki, 469). The program can be based on Support Vector Machines (SVMs), for the foundation of the algorithm. SVMs are supervised learning models that analyze data and proceed to decisions, based on their knowledge (Kampouraki, 467). Using this, the program can ask the patient for relevant information such as medical history and current symptoms to search for a reasonable diagnosis in its myriad of data sets.

A program fitted with machine learning AI has more potential power than that of a medical professional. The biggest benefit of a machine learning program is that it can easily handle large data sets with ease. This means that the program will be able to match up user data with existing data to efficiently compile a set of answers. In Leung’s article it is mentioned that one of the principle uses of machine learning is predicting phenotypes from biomarkers such as the genome (Leung, Section II). The algorithm is set to go through the data sets in strands of DNA in order to locate errors or diseases. No human would be able to do the same in a reasonable amount of time. As the processing power of computers begin to increase, the speed and efficiency will only increase. Leung and his team’s programming follows a powerful approach where the computational model is trained to predict measurable intermediate cell variables (Leung, Section II). This is a task a human would not be able to do, and shows the level of intelligence these programs can have. Extrapolating on this idea, it is not impossible to either go through a patient user’s DNA sample or to compare user submitted data, such as medical history and pictures, with pre-existing data to determine a reliable diagnosis.

One of the biggest challenges that must be overcome is the relationship and connection between software engineers and medical professionals. There is a stigma against engineers by doctors that do not believe that an automated program would ever be reliable enough to replace a physical doctor. Although this proposal is not meant to completely replace a physical doctor, the question of reliability still stands. An “algorithm” can reliably predict a possible disease that a patient may have.  A performance test done by Manickapriya and Nimala, researchers from the Department of Information Technology in SRM University, seems to suggest that they were able to validate the accuracy and efficient of active learning leading to reliable and authentic predictions (Manickapriya & Nimala, 2636). They state that clustering data sets are useful, but it is hard to determine the value of some preference “p” or how to measure similarity between user-submitted data and existing data sets (Manickapriya & Nimala, 2636). This test shows that the diagnosis done by the proposed program will most likely be reliable depending on the pre-existing data sets, and how the program is trained to learn new information. Another method to test for reliability is to label a new set of data and adding it to the learning set. Describing the check for reliability, Kononenko stated “If the decision does not vary a lot, we assume that the classifier is quite reliable [and] if the decisions are sensitive to adding a new case to the learning set, the final decision is not reliable.” (Kononenko, 4.1.1). In other words, by adding a new set and manually checking if the program had properly implemented and used that data, it can be made sure that the program is reliable.

A machine learning AI used to create an automated medical diagnosis would close the gap between many areas. Its cheap costs to implement relative to the costs of implementing a traveling team of doctors would allow for the program and medical care to be more widespread in areas that need it most. It would allow for the acceleration of medical knowledge in technology in areas that lack a basic medical foundation or facility, would allow for cheap implementation across the world, and would build a stronger connection between engineers and doctors that would allow for future enhancements in medicinal technology. The program would be able to help places such as the Central Africa Republic, as well as expanding the help towards other areas such as Syria and Afghanistan.  Overall, with the money saved by efficiently, but effectively, helping and supporting a country, a reasonable level of medical care and education can be brought to many more developing countries.

                                                                    Works Cited

"Central African Republic." MSF USA. Medicins Sans Frontieres, 20 Dec. 2013. Web. 08 Apr. 2016.

"Donation FAQ." MSF USA. N.p., 18 Nov. 2013. Web. 15 Mar. 2016.

International Crisis Group, comp. Central African Republic: The Roots of Violence. Rep. no. 230. N.p.: International Crisis Group, 2015. Print.

Kampouraki, A., Vassis, D., Belsis, P., & Skourlas, C. (2013). e-Doctor: A Web based Support Vector Machine for Automatic Medical Diagnosis. Procedia - Social And Behavioral Sciences, 73(Proceedings of the 2nd International Conference on Integrated Information (IC-ININFO 2012), Budapest, Hungary, August 30 - September 3, 2012), 467-474. doi:10.1016/j.sbspro.2013.02.078

Kononenko, I. (2001). Machine learning for medical diagnosis: history, state of the art and perspective. Artificial Intelligence in medicine, 23(1), 89-109.

Leung, M. K., Delong, A., Alipanahi, B., & Frey, B. J. (2015). Machine Learning in Genomic Medicine: A Review of Computational Problems and Data Sets. Proceedings Of The IEEE In Press, 4 December 2015, doi:10.1109/JPROC.2015.2494198

Manickapriya, S., & Nimala, K. (2015). Machine learning approach for medical diagnosis. ARPN Journal Of Engineering And Applied Sciences, 10(6), 2634-2637.

"Right to Healthcare around the Globe." Humanium - Together for Children's Rights. Humanium, n.d. Web.