Monday 03 March 2025
A novel approach has been developed to classify neonatal ultrasound videos into different viewpoints, a crucial step in diagnosing heart conditions in newborns. The technique, which combines computer vision and machine learning algorithms, has shown promising results in a recent study.
Ultrasound imaging is a widely used diagnostic tool in pediatrics, allowing doctors to visualize the internal organs of newborn babies. However, analyzing these images can be time-consuming and requires expertise. Automated classification of ultrasound videos could streamline this process, enabling faster diagnosis and treatment of cardiac conditions.
The researchers behind this new approach have developed a system that uses a type of neural network called a convolutional neural network (CNN) to classify the viewpoints in neonatal echocardiogram videos. The CNN is trained on a dataset of labeled images, which are then used to predict the viewpoint of unseen images.
In traditional image classification tasks, the input data is typically a single frame or still image. However, ultrasound videos are dynamic and contain information that changes over time. To capture this temporal information, the researchers developed a novel feature weaving approach that combines spatial and temporal features from multiple frames in the video.
The feature weaving approach involves dividing the output of the CNN into segments, which are then reconstructed to form new feature vectors. This process is repeated for each frame in the video, creating a sequence of feature vectors that can be fed into a recurrent neural network (RNN) for classification.
The RNN uses these feature vectors to predict the viewpoint of the image at each time step. The final class probabilities are taken from the last time step, providing the predicted viewpoint for the entire video.
In a recent study, the researchers tested their approach on a dataset of 1049 neonatal echocardiogram videos, which were labeled with one of 16 different viewpoints. The results showed that the feature weaving approach outperformed traditional image classification methods, achieving an accuracy of 93.8% and an F1-score of 93.7%.
These findings have important implications for the diagnosis and treatment of cardiac conditions in newborns. Automated classification of ultrasound videos could enable faster and more accurate diagnoses, allowing doctors to provide timely treatment and improve patient outcomes.
The researchers plan to further develop their approach by increasing the size of their dataset and exploring other techniques for feature extraction and classification. With continued advancements, this technology has the potential to revolutionize the field of pediatric cardiology, enabling earlier and more effective diagnosis and treatment of heart conditions in newborns.
Cite this article: “Automated Classification of Neonatal Ultrasound Videos for Cardiac Diagnosis”, The Science Archive, 2025.
Neonatal, Ultrasound, Classification, Machine Learning, Computer Vision, Neural Network, Cnn, Rnn, Pediatric Cardiology, Cardiac Conditions
