Artificial Intelligence (AI) through the use of computer algorithms and software, is transforming how the healthcare industry thinks about disease diagnosis, prevention, and treatment. Prior to the use of AI, medically trained professionals relied on a trained eye to understand and diagnose many different types of diseases. However, even the best radiologists were prone to misdiagnoses. The trained eye can now be complemented by applications to analyze possible relationships between prevention/treatment techniques and patient outcomes.
For the poster, my researched focused on CheXNet, a deep-learning algorithm developed by Stanford researchers to evaluate chest X-rays for signs of disease. CheXNet is comprised of a 121-layer convolutional neural network that takes a chest X-Ray image as an input, and outputs the probability of a pathology. The algorithm can diagnose up to 14 types of medical conditions.
The new technological advancement in AI is especially important when considering the amount of people that are affected by pneumonia each year. Detecting pneumonia in patients is difficult as X-ray images are often vague and overlap with other diseases. Below are two statistics stressing the significance in just the United States.
1 million adults are hospitalized.
50,000 die each year from the disease.
The objective of my research into CheXNet is to inform students of the evolving role AI is playing in diagnosing disease. Students will learn about:
How AI can assist humans and how humans can assist AI.
Facilitation of accelerated diagnosis & medication nonadherence.
Career outlook for students seeking entrance into the healthcare industry.
The Stanford researchers used two methods in the development of CheXNet.
Collected annotations from four practicing radiologists on a subset of 420 images from ChestX-ray14.
Made modifications to CheXNet to detect all 14 types of medical conditions in ChestX-ray14.
The researchers found in their first method that the model exceeded average radiologist performance on pneumonia detection. The results of the radiologists above and CheXNet were based upon the F1 Score. The F1 Score is a measure of accuracy combining both precision and recall in computing the score. At a score of 0.435 for CheXNet compared to 0.387 for the radiologists, the difference in results was statistically significant.
The researchers found in their second method that CheXNet achieves state of the art results on all 14 pathology classes. The algorithm outperformed all of the best published results on all 14 pathologies in the ChestX–ray14 dataset.
With AI technology like CheXNet, the hope is that medical imaging expertise can be brought to areas of the world where access to skilled radiologists is limited. After about one month, the Stanford researchers were able to outperform expert radiologists at diagnosing pneumonia. In conclusion, the algorithm helps reduce the number of missed cases of pneumonia, and by showing radiologists where to look first, treatment can begin much sooner for the sickest patients.
According to an article titled “3 ways artificial intelligence is changing the healthcare industry”, between 1988 and 1994, roughly 38 percent of adults living in the United States were taking at least one prescription drug. Later the number would increase to 49 percent. However, 3.2 billion prescriptions out of this 49 percent are not taken as directed or not taken at all. Artificial intelligence will be able to utilize algorithmic tools to identify which patients are most prone to medication nonadherence and cut out wasted prescriptions.
In the above graph from Accenture, artificial intelligence will be able to fill the gap in the disparity between the supply and demand of clinicians. Since growth in the AI health market is expected to reach $6.6 billion by 2021, a background in computer science could very well help students break into the medical field since must curriculums have not yet shifted from the information age to the age of artificial intelligence. Mckinsey estimated in 2017 that 50% of activities carried out by workers have the potential to be automated. Something to keep an eye on for professionals seeking a career in healthcare.