| Abstract [eng] |
Objective: Ultrasound is the first-line imaging modality in obstetrics due to its accessibility and real-time properties. Yet correct visualization of a fetus in utero with ultrasound imaging is highly operator-dependant and needs specialized training. To overcome human biases and the intra- and interobserver variability seen during the performance of obstetric ultrasound, artificial intelligence is currently being implemented into the sonographic assessment of fetuses. Due to its potential benefits and the capability to analyse large amounts of data, algorithms have already gained prominence and acceptance in medical ultrasound imaging fields, such as echocardiography. The objective of this thesis is to explore the advancement of obstetrical ultrasound and provide an overview of the current knowledge of artificial intelligence enhanced obstetrical ultrasound. Different algorithms will be reviewed for their efficacy, strengths and limitations including studies comparing the performance of artificial intelligence directly to human performance. Methods: A literature review was performed searching the PubMed and Google Scholar databases to identify studies about artificial intelligence and ultrasound in obstetrics. Articles were selected based on the eligibility criteria and analysed systematically in accordance with relevant related topics. Results: A total of 61 articles were included in this review. The literature indicates that ultrasound in obstetrics has progressively improved since its introduction over 60 years ago. Currently artificial intelligence is being implemented into sonographic practice to make examinations more consistent. This review demonstrates the functioning of artificial intelligence together with its efficacy and clinical utility in biometric and anatomical assessment at present. Comparative studies evaluating the performance of artificial intelligence against human operators show that algorithms are capable to generate expert level interpretations and are improving training efficiency, acting as a supportive tool. Conclusion: Based on the results a potential benefit of including artificial intelligence in obstetrical ultrasound is seen. Deep learning, which is performing particularly well in image pattern recognition makes ultrasound examinations faster and more reliable. The included studies show that high observer variabilities are decreasing and detection rates of abnormalities are facilitated. Artificial intelligence performs on an expert level which improves training effectiveness and performance of novices. This could lead to a potential benefit for underserved areas across the globe where the access to well-trained obstetric sonographers is limited. Yet, the included data is heterogenous in nature, which makes a meaningful comparison unpracticable. Furthermore, there are few prospective randomized controlled trials testing the real clinical utility of artificial intelligence in obstetric ultrasound. |
