Application of the U-Net Deep Learning Model for Segmenting Single-Photon Emission Computed Tomography Myocardial Perfusion Images

2024

Using U-Net Deep Learning for Heart Image Segmentation

Sample size: 1100 publication 10 minutes Evidence: high

Author Information

Author(s): Alenezi Ahmad, Mayya Ali, Alajmi Mahdi, Almutairi Wegdan, Alaradah Dana, Alhamad Hamad, Koulaouzidis George

Primary Institution: Kuwait University

Can the U-Net deep learning model improve the accuracy of segmenting myocardial perfusion images?

The U-Net model effectively improves segmentation accuracy in myocardial perfusion imaging, facilitating better diagnosis.

The U-Net model achieved a test accuracy of 98.9%.
The model's performance was comparable to expert nuclear medicine technologists.
A dataset of 4560 images and corresponding masks was generated for training.
High segmentation accuracy was demonstrated with a Dice score of 94%.
The model showed robustness across various evaluation metrics.
Data augmentation techniques were applied to enhance model performance.
The study utilized a large and diverse dataset to minimize selection bias.
The U-Net model outperformed other segmentation frameworks in the study.

This study shows that a computer program can help doctors see heart images more clearly, making it easier to find problems.

The study used a U-Net deep learning model trained on 1100 myocardial perfusion imaging studies to enhance segmentation accuracy.

Potential biases due to the retrospective nature of the study and reliance on expert labeling.

The study is limited by its retrospective design, which may introduce biases related to data collection.

Patients aged 17 to 88 years, with a diverse representation of healthy and unhealthy individuals.

0.1

1.005–1.017

p = 0.1

Access the complete publication on the publisher's website