Use of artificial intelligence as a didactic tool to improve echocardiography interpretation by non-cardiologists

Background:

Artificial-intelligence (AI) is used in different imaging modalities in medicine by non-cardiologists (NC) to aid in diagnosis and decision making. Incorporating AI into echocardiography operated by NC to accurately assess LV function (LVF) may lead to better diagnostic decisions.

The hypothesis of this study is that AI can be used as a didactic tool to improve NC assessment of LVF from the apical 4-chamber (A4ch) view.

Methods:

This is a prospective randomized controlled trial comparing the ability of 16 physicians and physician assistants to assess A4ch clips for LVF before and after introduction to AI.

Following a didactic course, participants were randomly divided into two groups (GrpA, GrpB). In the 1st session (S1) both groups were shown 20 clips and were asked to assess LVF and categorize it as normal, segmental dysfunction, or global dysfunction. Following each clip assessment, only one group (GrpA) was shown the results of an AI-based application (LVivo, DiA Imaging Analysis).

For the 2nd session (S2) both groups were presented with a different set of 40 clips and asked to evaluate the same values.

In order to reduce the effect of an assumed cumulative improvement on baseline assessment success, S1 was divided into 4 first clips in each LVF category (12 clips, S1.1) and last 8 clips (S1.2).

Results:

Demographic characteristics did not differ between two groups.

Comparing the success rate in S1.2 and S1.1, there was a trend toward higher delta and ratio in GrpA (Median[IQR]; Delta:0.29[0.15-0.42], 0.06[-0.02-0.24], respectively, P=0.105. Ratio: 1.68[1.26-2.25], 1.09[0.94-1.45], respectively, P=0.083).

Comparing the success rate in S2 and S1.1, the delta and the ratio were higher in GrpA (Median[IQR]; Delta:0.11[-0.5-0.3], 0.01[-0.5-0.14], respectively, P=0.234. Ratio: 1.30[0.92-1.91], 1.03[0.91-1.25], respectively, P=0.234).

Conclusion:

The introduction of medical personnel to AI showed a cumulative trend of improvement in the success rate of LVF assessment.