Machine Learning Algorithms for Classification of First‐Trimester Fetal Brain Ultrasound Images

Abstract

Objective-To evaluate the feasibility of machine learning (ML) tools for segmenting and classifying first-trimester fetal brain ultrasound images.Methods-Two image segmentation methods processed high-resolution fetal brain images obtained during the nuchal translucency scan: "Statistical Region Merging" (SRM) and "Trainable Weka Segmentation" (TWS), with training and testing sets in the latter. Measurement of the fetal cerebral cortex in original and processed images served to evaluate the performance of the algorithms. Mean absolute percentage error (MAPE) was used as an accuracy index of the segmentation processing.Results-The SRM plugin revealed a total MAPE of 1.71% AE 1.62 SD (standard deviation) and a MAPE of 1.4% AE 1.32 SD and 2.72% AE 2.21 SD for the normal and increased NT groups, respectively. The TWS plugin displayed a MAPE of 1.71% AE 0.59 SD (testing set). There were no significant differences between the training and testing sets after 5-fold cross-validation. The images obtained from normal NT fetuses and increased NT fetuses revealed a MAPE of 1.52% AE 1.02 SD and 2.63% AE 1.98 SD.Conclusions-Our study demonstrates the feasibility of using ML algorithms to classify first-trimester fetal brain ultrasound images and lay the foundation for earlier diagnosis of fetal brain abnormalities.

Computational methods for the analysis of early-pregnancy brain ultrasonography: a systematic review

“…16 (29%) studies reported the computational time, and 7 (13%) studies reported what kind of computational hardware was used. 23 , 24 , 25 , 26 , 27 , 28 , 29 , 30 , 31 , 32 , 33 , 34 , 35 , 36 , 37 , 38 , 39 , 40 , 41 , 42 , 43 Computation time ranged from 70 micro seconds to 25 min. 32 , 35 Next, we discuss the studies per topic in the following order: biometry, standard plane detection, segmentation, and together: growth models, abnormality detection, quality enhancement and visualization.…”

“…The fetal head in the TV plane was segmented by five studies. 38 , 40 , 68 , 69 , 70 Three of those used the aforementioned HC18 challenge dataset. 40 , 44 , 68 , 69 Furthermore, four studies segmented the cerebellum, 39 , 41 , 71 , 72 and three studies segmented the choroid plexus.…”

“… 40 , 44 , 68 , 69 Furthermore, four studies segmented the cerebellum, 39 , 41 , 71 , 72 and three studies segmented the choroid plexus. 38 , 39 , 72 All these studies used a learning-based method, and four studies used data acquired in the first trimester (<14 weeks). 38 , 40 , 68 , 69 Fig.…”

A Machine Learning Method for Automated Description and Workflow Analysis of First Trimester Ultrasound Scans

“…In medical image analysis, a number of studies have explored video classification and analysis methods for secondtrimester fetal ultrasound anomaly scan [10]. These include image-based segmentation [11]- [14], frame classification [15], [16], fetal biometry [17], [18] and tracking [19]. Several recent studies have explored obstetric ultrasound standard plane detection [20]- [23], automated biometry [13], [18], [24], [25], activity captioning [26]- [28] and visual attention modelling [20].…”

“…In contrast, only a few first trimester US image analysis studies have been undertaken with a different focus: automated CRL and NT measurements [38], assessment of the maternal placenta [39], classification of fetal brain images [40] and fetal echocardiography [41]. As exemplars of the state of art, Mathewlynn et al [42] proposed a fully automated placental volume and vascularity measurements method.…”