The efficiency of a Machine learning approach based on Spatio-Temporal information in the detection of patent foramen ovale from contrast transthoracic echocardiography Images: A primary study

Objectives: As a noninvasive method, diagnostic rate of PFO by CTTE is controversial because of its subjectivity and the examiner's dependence about the bubbles identification. In order to improve the diagnostic accuracy, a machine learning algorithm diagnostic method for PFO based on CTTE images (AI-CTTE) was established in our primary study. This study aimed to investigate the efficiency of the AI method in PFO detect,and to clarify the clinical value of AI in identifying PFO.Methods: From August 2019 to June 2021, 200 patients with suspected PFO in six hospitals were eventually enrolled, and all the patients underwent CTEE and CTTE examinations respectively. Each case contained ul-trasound videos of 5 cardiac cycles before and after injecting agitated saline with Valsalva maneuver from apical four-chamber (A4C) view. The patients with or without PFO were identified by the machine learning algorithm and other non-operating physicians (including experts and junior physicians). Results: A total of 144 out of 200 (72%) patients were confirmed PFO by CTEE. CTEE was used as a gold standard to assess the model's diagnostic valve.The sensitivity and specificity of the AI-CTTE for PFO detection were 78% and 75% respectively. In comparison, the sensitivity and specificity of the ultrasonic experts group were 86% and 89%, while those of the ultrasound residents group were 62% and 82%. The diagnostic efficiency of the AI-CTTE was similar with the ultrasonic experts group, but significantly higher than the ultrasound residents group.Conclusions: As a non-invasive method, AI can provide diagnostic help for junior physicians, it creates a new way for PFO recognition.