- Journal
- BIOMEDICAL SIGNAL PROCESSING AND CONTROL
- Année
- 2025
- Volume
- 104
- Article
- 107493
- Mois
- JUN
- DOI
- 10.1016/j.bspc.2025.107493
Abstract
Recent advances in Generative Artificial Intelligence (GAI) have revolutionized various fields, notably medical imaging. In the context of Parathyroid Glands detection in Nuclear Medicine, physicians rely on a subtraction process involving two different images. Consequently, the subtracted image depends entirely on the two images. This methodology is manual, time-consuming, and depends entirely on medical expertise. Conditional GAI, a sub-field of GAI, generates new data based on input specifications, enabling more controlled results. However, generative models require a large amount of data and still present limitations in precisely controlling outputs over inputs. This paper presents anew model named Nuclear Medicine U-Net Siamese Network (NM-USNet) based on a combination of U-Net and Siamese networks. The U-Net takes two input images and generates one, while the Siamese network distinguishes between the generated and the reference images. The proposed methodology achieves a mean correlation of 0.956 compared to reference images, showing its potential as a reliable medical tool to assist physicians in the detection of parathyroid glands. This advancement not only improves the accuracy of diagnosis, but also represents a significant step forward in integrating GAI into medical practice. The proposed approach gives high results despite the limited data size.