Publications by authors named "Donato Zigrino"
J Imaging Inform Med
October 2024
Article Synopsis
- - Medical image classification using convolutional neural networks (CNNs) usually needs a lot of manual adjustments, but neural architecture search (NAS) can automate this process, making it more efficient.
- - This study focuses on classifying different sub-types of cardiac amyloidosis using NAS on [18F]-Florbetaben PET cardiac images, with a significant dataset augmentation from 597 to 4048 images, resulting in 5000 evaluated architectures.
- - The best-performing network, named NAS-Net, achieved 76.95% overall accuracy and showed impressive sensitivity and specificity rates for the AL and ATTR-CA subjects, confirming that NAS can rival traditional methods while using fewer parameters.
