While artificial intelligence can improve upon state-of-the-art diagnostic procedures, such as for detection of breast cancer, it can also promote the broader use of new diagnostic techniques. Recent research published in JAMA Cardiology illustrates how AI can assist with lung ultrasound (LUS) procedures for the diagnosis, assessment, and monitoring of patients with shortness of breath.
AI in Lung Ultrasound Procedures
LUS is a portable, low-cost alternative to traditional chest radiography, eliminating radiation exposure while providing real-time and point-of-care assessments. Detection accuracy of LUS can be better than radiography for pneumonia, pneumothorax and other lung pathologies, but only with sufficient procedure training. Acquiring high-quality LUS images currently relies on sonography experts with advanced technical skills and specific LUS experience. The need for such expertise limits the broad use of LUS, especially in underserved areas that lack access to these experts. To address this issue, Lung Guidance AI software (Caption Guidance™, GE HealthCare) was developed to help trained healthcare professionals (THCPs) such as medical assistants, registered nurses and physicians to acquire high-quality LUS images without the need for specific LUS training.
Researchers compared the results of LUS procedures performed by LUS experts on patients with suspected pulmonary edema to those performed by THCPs trained only on the AI software. Procedures took place at an outpatient site, an inpatient site, and at two emergency departments, for a total of 176 patients. The study found that 98.3% of the THCP examinations provided diagnostic-quality images and that there were no statistically significant differences in the study-level image quality between the two groups (P > 0.05). While the authors recommended that more studies be performed to expand the software validation to other lung pathologies, these results support the use of AI-assisted LUS for evaluation of pulmonary edema by those with no prior LUS experience.
The FDA
Of course, this isn’t the end of the story for those of us who work to get new technologies onto the market. The US FDA has just issued a new draft guidance document on the use of AI-enabled device software functions. Fortunately, the guidance holds no surprises and promotes the usual verification and validation studies on user interface and performance, as well as monitoring and cybersecurity considerations.
The idea that AI can transform health care by assisting health care providers and improving patient care is not new, of course. This study on AI-assisted LUS, and the new guidance by FDA, adds more evidence that AI has a useful place in medical device design.