Computer-Aided Detection and Segmentation System of lesions of COVID-19 and Community-Acquired Pneumonia and their extension in CT lung images

• Ano de defesa: 2023
• Autor(a) principal: Motta, Pedro Crosara
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Palavras-chave em Português: COVID-19; Sistema de auxilio ao diagnóstico por computador; CNN; Segmentação; Classificação; Imagens médicas; exame TC; Validação externa; Computer Aided Diagnostic; Segmentation; Classification; Medical image; CT Scan; External validation; CNPQ::ENGENHARIAS::ENGENHARIA ELETRICA
• Link de acesso: http://repositorio.ufc.br/handle/riufc/74548

Even with more than 80\% of the population wholly vaccinated for COVID-19, the disease still claims victims. Thus, having a Computer Aided Diagnostic system that can securely assist in identifying COVID-19 and determining the level of care required and if the disease is progressing or digressing, particularly in the Intensive Care Unit, is crucial in the fight against this epidemic. To create such tool, we first merged public datasets from the literature to train Lung and Lesion segmentation models from different distributions. Then we trained eight CNN models for COVID-19 and Common Acquired Pneumonia classification. Finally, if the exam is classified as COVID-19, we quantified the lesions and evaluated the severity of the full CT Scan. For external validation on SPGC Dataset, using Resnext101 Unet++ and MobileNet Unet for lung and lesion segmentation, respectively, we achieved an accuracy of 98.05%, an F1-score of 98.70\%, a precision of 98.7%, a recall of 98.7%, and a specificity of 96.05\%, needing only 19.70 seconds per full CT scan. Finally, when classifying these detected lesions, Densenet201 reached an accuracy of 90.47%, an F1-score of 93.85\%, a precision of 88.42\%, a recall of 100.0\%, and a specificity of 65.07%. The results showed that our pipeline correctly detected and segmented lesions from COVID-19 and Common Acquired Pneumonia in CT scans, differentiating these two classes from Normal exams.