Use of real‐time artificial intelligence in detection of abnormal image patterns in standard sonographic reference planes in screening for fetal intracranial malformations

Abstract

What are the novel findings of this work?In this study, we developed and validated an artificial intelligence system, the Prenatal ultrasound diagnosis Artificial Intelligence Conduct System (PAICS), to detect nine specific intracranial-malformation patterns in standard sonographic reference planes of the fetal central nervous system (CNS). The PAICS achieved excellent performance on both internal and external validation, with accuracy comparable to that of expert sonologists, while requiring significantly less time. What are the clinical implications of this work?The PAICS is a real-time artificial intelligence-aided image recognition system capable of detecting fetal intracranial malformations. This fast, accurate algorithm has the potential to be an effective and efficient tool in screening for congenital CNS malformations. It should be particularly useful in community hospitals, which often lack imaging expertise.

Real‐time identification of fetal anomalies on ultrasound using artificial intelligence: what's next?

“…To assess the effect, risks and clinical value of algorithms, Bitterman et al 5 suggest a five-level model for medical AI autonomy (Figure 2). According to this model, the real-time AI system of Lin et al 2 should be classified as Level 2: clinical decision-support. This means that, although the system shows a remarkable leap towards autonomy in the detection of abnormalities, it is only a tool, leaving the clinician not only responsible but also liable for delivering correct diagnoses.…”

“…Fast‐forward to the current issue of the Journal, in which Lin et al 2 . present a real‐time AI system that can identify automatically standard reference planes of the fetal brain (transthalamic, transventricular and transcerebellar) during a routine ultrasound scan.…”

“…suggest a five‐level model for medical AI autonomy (Figure 2). According to this model, the real‐time AI system of Lin et al 2 . should be classified as Level 2: clinical decision‐support.…”

“…This process can be thought of as the 'sonographer loop' 3 : the sonographer moves the transducer, looking for a standard plane, they receive real-time visual feedback from the video on the display screen and respond, constantly moving their hand to adjust the transducer position, which changes the displayed video. Reflecting on this feedback loop, the system presented by Lin et al 2 involves two components that should be considered separately. The first is automatic identification of standard planes and the requisite anatomical landmarks, all of which are displayed in real time on the screen.…”

“…Fast-forward to the current issue of the Journal, in which Lin et al 2 present a real-time AI system that can identify automatically standard reference planes of the fetal brain (transthalamic, transventricular and transcerebellar) during a routine ultrasound scan. In addition, the system identifies the salient anatomical landmarks in these planes and classifies each plane as normal or as presenting one of nine different intracranial anomalies.…”