Efficient and visualizable convolutional neural networks for COVID-19 classification using Chest CT
作者：
Highlights：
• 20 CNNs were trained and evaluated for COVID-19 Classification from Chest-CT scans.
• EfficientNetB5 attained highest accuracy(0.976), sensitivity(0.979), & F1 score (0.977).
• Model generalizes to new datasets & predicts with ∼ 0.1 s/img on GPU & ∼0.5 s on CPU.
• GradCAMs localized ground-glass opacities and consolidations.
• Intermediate Activation Maps demonstrated the model’s learning behavior.
摘要
•20 CNNs were trained and evaluated for COVID-19 Classification from Chest-CT scans.•EfficientNetB5 attained highest accuracy(0.976), sensitivity(0.979), & F1 score (0.977).•Model generalizes to new datasets & predicts with ∼ 0.1 s/img on GPU & ∼0.5 s on CPU.•GradCAMs localized ground-glass opacities and consolidations.•Intermediate Activation Maps demonstrated the model’s learning behavior.
论文关键词：Computed tomography,Convolutional neural networks,COVID-19,Deep learning,EfficientNets
论文评审过程：Received 21 April 2021, Revised 17 August 2021, Accepted 10 January 2022, Available online 20 January 2022, Version of Record 5 February 2022.
论文官网地址：https://doi.org/10.1016/j.eswa.2022.116540
