The most frequent chromosomal anomaly causing intellectual impairment and developmental delay is Down syndrome, also known as Trisomy 21, which can be detected in pregnancy. Many expectant mothers want to know if their fetus has this condition.
Visualization of the response region to show class-specific information for multilevel feature. Image Credits: Institute of Automation of the Chinese Academy of Sciences (CASIA)
To execute a non-invasive Down syndrome screening using ultrasound images, researchers at the Institute of Automation of the Chinese Academy of Sciences (CASIA) have now built an intelligent prediction model.
On June 21st, 2022, the study appeared in JAMA Network Open.
Due to the technique’s reliability, ease of use, and low cost, ultrasound scans have been regularly utilized for decades to screen fetuses for Down Syndrome. But in actual ultrasound tests, detection accuracy is less than 80% using standard ultrasound indications.
The detection of Down syndrome frequently involves invasive techniques such as villus biopsy, amniocentesis, and fetal umbilical venipuncture.
In this study, the researchers built a deep learning (DL) model using a convolutional neural network (CNN) to learn representative characteristics from ultrasound pictures in order to detect fetuses with Down syndrome.
A CNN is a deep learning system that can take an input image, prioritize distinct elements and objects within it (i.e., apply biases and learnable weights), and distinguish between them.
Several hidden layers could be found in a CNN. The first layer picks up on edge detection, while the last layer picks up on more intricate form detection. There were eleven covert layers in this study.
The researchers also employed a class activation map (CAM) to clarify what the model concentrated on and how it directly allowed the CNN to acquire discriminative features for risk ratings, further interpreting the DL model in a human-readable way.
Between 11 and 14 weeks of gestation, the fetal face’s midsagittal plane was imaged using two-dimensional ultrasound technology. Each picture was divided into sections using a bounding box to display just the fetal head. The study included 272 people in the validation set and 550 participants in the training set, for a total of 822 cases and controls.
The first five tiers of feature maps produced by CAM, according to the researchers, vividly depicted the process of learning representative characteristics. The last layer’s CAM revealed the visually represented response zones 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 labeling markers, improving prediction accuracy by more than 15%. It is also superior to the current conventional invasive screening method for Down Syndrome based on maternal serum.
Jie Tian, Study Corresponding Author, Institute of Automation of the Chinese Academy of Sciences (CASIA)
The suggested concept is projected to develop into a non-invasive, affordable, and practical Down syndrome early pregnancy screening tool.
The National Natural Science Foundation of China and the Key R&D Program of the Ministry of Science and Technology provided funding for the study.
Journal Reference:
Zhang, L., 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. doi:10.1001/jamanetworkopen.2022.17854