Neural Network Sagittal Plane Detection in Prenatal Ultrasound
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Ultrasound images for prenatal diagnosis, particularly in the first trimester, often suffer from noise and boundary blur, making it difficult for professionals to accurately measure fetal parameters like the zona pellucida thickness on the middle sagittal plane, which is crucial for detecting Down's syndrome and other fetal defects.
Innovation Solution
A neural network system is developed to automatically detect the sagittal plane in 3-D medical images, using a convolutional neural network to generate a prediction sagittal mask and adjust parameters based on loss function data, enabling accurate identification of the middle sagittal plane from ultrasound images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual measurement by professional clinical personnel is used, then measurement capability is available, but time consumption and human error increase
Solution Approach 1:
The system enables self-service measurement by automatically detecting the middle sagittal plane and calculating fetal parameters without requiring continuous manual intervention. The neural network autonomously processes ultrasound images to identify anatomical landmarks and compute measurements, replacing the need for persistent human analysis while maintaining measurement capability.
Solution Approach 2:
The patent replaces the mechanical/manual measurement process with an automated computational system. A neural network model processes ultrasound images to automatically detect the middle sagittal plane and calculate fetal parameters, substituting the manual mechanical measurement process with an automated digital system that reduces both time consumption and human error.
2Measurement precision
If manual detection of middle sagittal plane is performed, then correct measurement plane can be identified, but the process becomes time-consuming and difficult
Solution Approach 1:
The patent introduces an auxiliary detection mechanism that uses readily identifiable anatomical landmarks (such as the fetal head and spine) as intermediaries to locate the middle sagittal plane. The neural network detects these prominent features first, then uses them as reference points to accurately identify the target measurement plane, simplifying the overall detection process.
Solution Approach 2:
The system performs preliminary detection of prominent anatomical landmarks and structural features before attempting to identify the middle sagittal plane. By pre-identifying key reference points such as the fetal head contour and spinal column, the system prepares the necessary information in advance, making the subsequent plane detection more straightforward and accurate.
3Measurement precision
If traditional ultrasound image analysis is used, then fetal parameters can be measured, but boundary blur and noise reduce measurement accuracy
Solution Approach 1:
The patent replaces traditional manual image analysis with an automated neural network-based system that is less susceptible to noise and boundary blur. The deep learning model processes the ultrasound images to automatically identify anatomical structures and measurement points, reducing the impact of image quality issues on measurement accuracy.
Solution Approach 2:
The system creates an enhanced digital representation of the fetal anatomy by processing the ultrasound image through a neural network. This computational model generates a refined interpretation of the image data, effectively creating a cleaner, more accurate representation that reduces the impact of noise and boundary blur on measurement precision.
Data Source
AI summary
A method of training neural network for obtaining medical sagittal image includes: using a first neural network on a 3-D medical image to generate a prediction sagittal mask; generating a prediction result according to the 3-D medical image and the prediction sagittal mask; generating a ground truth result according to the 3-D medical image and a ground truth sagittal mask; using a second neural network on the prediction result and the ground truth result; generating a loss function data according to an output of the second neural network; and adjusting parameters of the first neural network or the second neural network according to the loss function data.