Tecnologias da informação e métodos computacionais para gerenciamento, otimização e medicina de precisão em departamentos de imagens médicas

Abstract

The healthcare industry demands information technologies and computational methods to improve productivity and offer personalized assistance to patients. This work aimed to develop and explore information systems and computational methods to improve the management, optimize the exams, and provide new biomarkers and artificial intelligence signatures for decision making in medical imaging facilities. We developed softwares based on Workflow Based Approach (WBA) concept and computational methods using Python language to improve the management and optimize the exam protocols. A framework to access new biomarkers and artificial intelligence (AI) signatures was developed and validated in COVID-19 CT patients, 18F-FDG-PET/CT cervical cancer and 18F-FDG-PET/CT Hodgkin lymphoma patients. This framework was feasible for robustness analysis: repetitivity (error < 5%), reproducibility (intraclass correlation coefficient, ICC > 90%) and clinical correlation (p < 0.05). The overall performance of predictive models was AUC=0.74 and AUC=0.96 for 18F-FDG-PET/CT cervical cancer and 18F-FDG-PET/CT Hodgkin lymphoma, respectively. A new AI software to support thorax CT COVID-19 diagnoses was implemented and validated within PACS/Viewer. Without the support of software, physicians performed with mean sensitivity and specificity of 83.4% and 64.3%, respectively. When they were assisted with AI software, mean sensitivity and specificity were 87.1% and 91.1%, respectively. In addition, AI software improved the inter-rater reliability from moderate to substantial agreement in a Cohen’s Kappa scale.