Spina bifida: an algorithm anticipates diagnosis in the first trimester of pregnancy

An ultrasound scan performed in the first trimester of pregnancy can already tell a lot about the health of the fetus' brain. But today, with the help of artificial intelligence, it can do so very precisely by detecting cases of open spina bifida (OSB) and Dandy-Walker malformation, among the most challenging congenital abnormalities of the central nervous system. The credit for this goes to an international team, led by the Catholic University of the Sacred Heart in Rome with the Policlino Gemelli Irccs, which has developed a deep learning algorithm capable of automatically analysing the posterior region of the fetal brain in ultrasound scans taken between 11 and 14 weeks of pregnancy.

The results of the study involving 10 fetal medicine centres were published in the journal Radiology Artificial Intelligence.

"The early detection of these conditions," comments Professor Tullio Ghi, Professor of Gynaecology and Obstetrics at the Catholic University of the Sacred Heart in Rome and Director of the UOC Obstetrics and Obstetric Pathology at the Policlinico Gemelli Irccs Foundation, "allows patients to be sent promptly to a fetal medicine reference centre to offer them a more accurate diagnostic picture and adequate counselling by the end of the first trimester. This is of particular importance for open spina bifida, which is now amenable to prenatal surgical therapy in utero with significant improvements in clinical outcomes. Earlier ultrasound diagnosis makes it possible to plan individualised care of the child, perform thorough examinations (genetic tests and MRI) and carefully select cases that can benefit'.

'The study retrospectively analysed 251 ultrasound images of the fetal brain in the first trimester: 150 normal cases and 101 with abnormalities,' recalls Alessandra Familiari, Associate Professor of Gynaecology and Obstetrics at the Catholic University, UOC of Obstetric Pathology at Policlinico Gemelli, co-PI and first name of the study. The images were used to 'teach' the algorithm to recognise these abnormalities in the brain region of interest. Never before has artificial intelligence been successfully employed in the ultrasound study of fetal anatomy at such an early stage of pregnancy. Due to the innovative potential of this project, the study has been selected by the Ministry of Health as worthy of funding in the 2022 finalised research calls'.

"Our algorithm," says Luca Boldrini, researcher Diagnostica per Immagini e Radioterapia Università Cattolica del Sacro Cuore and Head of the UOS MR-guided external beam radiotherapy at Policlino Gemelli, "achieved an accuracy of 88% in distinguishing normal from pathological images, a value that indicates a high diagnostic capacity. Open spina bifida was identified with even higher accuracy (93%) and high sensitivity. This is a great result, considering that it uses routine ultrasound images, which are completely non-invasive for the foetus and the mother'.In other words, the well-trained algorithm was able to correctly recognise most pathological cases, reducing the risk of false negatives and false positives. The plus is that all this happened at a very early stage (between the 11th and 14th week of pregnancy), thus moving the hands of early diagnosis significantly backwards (these CNS abnormalities are currently diagnosed in the second trimester of pregnancy).