Algorithms – A Matter of Life and Death

TISB’s Grade 12 student, Anirudh, has presented a paper he authored at the IEEE conference, ‘ICADEE 2020’. The paper, titled ‘Gap Analysis of the Accuracy of Doctors versus Machine Learning Models for Pneumonia Detection from X-Rays’ was selected to be published, and to be part of the online research symposium, where some of the best research papers in the field of engineering and computer science are featured. It has now been published on IEEE Explore, an internationally reputed journal for papers on electrical engineering.

Current machine learning algorithms diagnose at an accuracy of about 90%. As diagnosis often can be a matter of life and death – Anirudh, Grade 12

After attending a machine learning course in the summer of 2019, Anirudh became fascinated by the potential impact that machine learning can have on medicine and healthcare. He therefore conducted a research study between the months of August and October in 2019, after which he wrote the paper in the month of December. 

“The paper is a comparison between the accuracy of various machine learning algorithms on pneumonia diagnosis, versus the accuracy of doctors when detecting pneumonia”, says Anirudh, who carried out the study by designing algorithms that could detect pneumonia from chest x-rays. “Essentially, I trained multiple computer algorithms to diagnose a patient for pneumonia based on chest x-rays”, explains Anirudh, who then hand-picked a sample of images and showed them to a sample of doctors, asking them to diagnose the same. As a final step he carried out a research on what aspects would aid in improving the accuracy of current algorithms and posed a solution to help automated disease diagnosis.

The main target audience of Anirudh’s research and paper are engineers around the world, specifically those working on image-processing. However, it may also be of use to hospitals that may see this as an opportunity to invest in and improve automated diagnosis. Current machine learning algorithms diagnose at an accuracy of about 90%. “As diagnosis often can be a matter of life and death, this accuracy rate is simply not high enough”, says Anirudh who believe that hospitals and healthcare professionals are interested in harnessing this type of technology in the near future. As radiologists are doctors that are scarce in numbers and stretched thinly, the need for automated diagnostic technologies is very high. As Anirudh’s paper address methods to improve current algorithm accuracy, “it is essential to understand how new algorithms should be designed, so that we can improve the accuracy of automated detection further – therefore lowering the cost of healthcare”, concludes Anirudh.

For more information on the IEEE conference, visit the ‘ICADEE 2020’ website.