The developed convolutional neural network would provide a cheap, convenient, and non-invasive way to screen for Down Syndrome in early pregnancy.
Trisomy 21, more commonly known as Down Syndrome, is the most common chromosomal abnormality that causes developmental delay and intellectual disability. This genetic disorder can be identified in utero and many pregnant women seek to determine whether their fetus has this abnormality.
For years, ultrasound images have been used to screen fetuses for Down Syndrome as it was cheap, convenient, and safe. However, with common ultrasound indicators, the detection accuracy is less than 80 per cent in actual ultrasound examinations.
Currently, while analysing cell-free fetal DNA in screening for Down Syndrome has shown a high accuracy (~99 per cent), some studies have suggested that more cost-saving approaches should be explored as such cell-free fetal DNA testing can be expensive.
Given the rising popularity of artificial intelligence in medical imaging analysis, a team of researchers from the Institute of Automation of the Chinese Academy of Sciences (CASIA) developed an intelligent prediction model to achieve the non-invasive screening of Down Syndrome using ultrasound images.
To do this, they developed a convolutional neural network to construct a deep learning model that could learn representative features from ultrasound images to pick out fetuses with Down Syndrome.
A convolutional neural network is a deep learning algorithm that takes an input image, assigns importance (i.e. learnable weights and biases) to various aspects/objects within the image and differentiates one from the other. Such a network can have as few as ten or as many as hundreds of hidden layers. The first layer learns how to distinguish edges and the last one learns how to make out more complex shapes. The team’s developed convolutional neural network involved 11 hidden layers.
To better make sense of the deep learning model in a human-readable form, the team also used a class activation map (CAM) to reveal what the model focused on and how it explicitly enabled the neural network to learn discriminative features for risk scores.
Here, the researchers examined two-dimensional images of the midsagittal plane of the fetal face between 11 and 14 weeks of gestation. Each image was segmented with a bounding box to show only the fetal head. A total of 822 participants were enrolled in this study, with 550 participants in the training set and 272 participants in the validation set.
From the results, the researchers found that the first five levels of feature maps revealed by CAM showed the process of learning representative features. The CAM applied to the final layer showed the visualised response regions for the model’s decision-making.
“This non-invasive screening model constructed for Down Syndrome in early pregnancy is significantly superior to existing, commonly used manual labelling markers, improving prediction accuracy by more than 15 per cent. It’s also superior to the current conventional invasive screening method for Down Syndrome based on maternal serum,” said Tian Jie, corresponding author of the study.
The team’s proposed model is expected to become a convenient, inexpensive, and non-invasive screening tool for Down Syndrome in early pregnancy.
Source: Zhang et al. (2022). Development and Validation of a Deep Learning Model to Screen for Trisomy 21 During the First Trimester From Nuchal Ultrasonographic Images. JAMA Network Open, 5(6), e2217854-e2217854.