Background: Convolutional Neural Networks (CNNs) are popular in every field of knowledge where the task is related to computer vision. Applications in Radiological imaging [1] are very spread, due to their capability to adapt to different problems. In this work we explored the ability of CNNs to classify X-rays images of patients with COVID-19, leveraging on three different architectures, nominally VGG16, Reset152 and Glyphnet [2]. Material and Methods: Networks were trained using a dataset composed by thousands of images of COVID-19, Viral Pneumonia and Normal Chest X-Rays, built by merging two large online datasets [3-5] with our dataset, available thanks to the Optimised project (Bando Ricerca Covid 19, Regione Toscana). Grad-CAM [6] was implemented to obtain the class activation heatmap for the image classification model. Preliminary Results: Results show an accuracy up to 98% for VGG16 and 84% for Reset152 (both leveraging on the transfer learning paradigm), whereas Glyphnet trained from scratch reached 84%. This work demonstrates the power of Deep CNNs in the task of classification of X-rays images of patient with COVID-19, moving towards a clinical implementation of such Artificial Intelligence based tools.
CONVOLUTIONAL NEURAL NETWORKS FOR CLASSIFICATION OF COVID19 X-RAYS IMAGES / Zini, C.; Ciacci, G.; Argenti, F.; Marzi, C.; Colantonio, S.; Buongiorno, R.; Romei, C.; Carpi, R.; Barucci, A.. - In: PHYSICA MEDICA. - ISSN 1120-1797. - ELETTRONICO. - 115:(2023), pp. 0-0. [10.1016/j.ejmp.2023.102881]
CONVOLUTIONAL NEURAL NETWORKS FOR CLASSIFICATION OF COVID19 X-RAYS IMAGES
Argenti, F.;Marzi, C.;
2023
Abstract
Background: Convolutional Neural Networks (CNNs) are popular in every field of knowledge where the task is related to computer vision. Applications in Radiological imaging [1] are very spread, due to their capability to adapt to different problems. In this work we explored the ability of CNNs to classify X-rays images of patients with COVID-19, leveraging on three different architectures, nominally VGG16, Reset152 and Glyphnet [2]. Material and Methods: Networks were trained using a dataset composed by thousands of images of COVID-19, Viral Pneumonia and Normal Chest X-Rays, built by merging two large online datasets [3-5] with our dataset, available thanks to the Optimised project (Bando Ricerca Covid 19, Regione Toscana). Grad-CAM [6] was implemented to obtain the class activation heatmap for the image classification model. Preliminary Results: Results show an accuracy up to 98% for VGG16 and 84% for Reset152 (both leveraging on the transfer learning paradigm), whereas Glyphnet trained from scratch reached 84%. This work demonstrates the power of Deep CNNs in the task of classification of X-rays images of patient with COVID-19, moving towards a clinical implementation of such Artificial Intelligence based tools.I documenti in FLORE sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.