AI Diagnoses Fractures and Assists in Treating Sepsis
When AI are supplied with data from thousands of doctors, they become more informed than any person could ever be. According to an article in Ars Technica, two recent studies have shown encouraging results that AI is capable of making decisions in diagnosis and treatment. Because the two studies are so different, it points to the variety of uses AI may one day have.
The first study sought to improve the guidelines for treating sepsis with a reinforcement learning algorithm. There was a large amount of data to evaluate: over 17,000 intensive care unit patients and 79,000 general hospital admissions from over 125 hospitals. Each patient included 48 bits of information, ranging from vital signs to lab tests to demographics. The AI Clinician, as it was named, recommended treatments that had the lowest mortality rate among the group of patients.
The Artificial Intelligence Clinician learns optimal treatment strategies for sepsis in intensive care
A reinforcement learning agent, the AI Clinician, can assist physicians by providing individualized and clinically interpretable treatment decisions to improve patient outcomes.
In the second study, the researchers wanted the AI to be able to identify hairline fractures, which are difficult to distinguish. Most instances of fractures need to be diagnosed in the emergency room and they hoped an AI would be able to assist non-specialists in this area.
Deep neural network improves fracture detection by clinicians
Historically, computer-assisted detection (CAD) in radiology has failed to achieve improvements in diagnostic accuracy, decreasing clinician sensitivity and leading to unnecessary further diagnostic tests. With the advent of deep learning approaches to CAD, there is great excitement about its application to medicine, yet there is little evidence demonstrating improved diagnostic accuracy in clinically-relevant applications. We trained a deep learning model to detect fractures on radiographs with a diagnostic accuracy similar to that of senior subspecialized orthopedic surgeons. We demonstrate that when emergency medicine clinicians are provided with the assistance of the trained model, their ability to accurately detect fractures significantly improves.
The study used data from over 135,000 images of possible wrist fractures to train their deep-learning convolutional neural network. They designed the algorithm to point out areas that would be most likely to have a fracture. The result was that accuracy in identifying fractures increased and false positives decreased. Using this algorithm would cut the rate of misdiagnosis in the ER nearly in half.
The next step for healthcare professionals is to integrate AI into their practice. Soon, clinical trials are expected to show solid numbers in diagnosis and treatment.
Reality Changing Observations:
Q1. What checks can we place on AI concerning the potential to make erroneous decisions?
Q2. Who is responsible when AI is used to support decision-making?
Q3. What effects would this AI have on the roles and skill-requirements of healthcare professionals?