The digitization of histological slides favors the migration of diagnostic practice to a fully digital environment. In this context, computer-aided diagnosis (CAD), with a focus on quantitative analysis, opens paths for diagnosis through artificial intelligence (AI) driven by technological development that favors strategies based on neural networks that uses images in this process. This study aims to develop and evaluate Deep Learning (DL) models to support the diagnosis of C&P diseases through digital histological analysis and clinical images. The research will be carried out at the Faculty of Medicine of the University of São Paulo - FMUSP (São Paulo, São Paulo, Brazil) and the sample will be obtained from different centers with the collaboration and support of Brazilian and foreign institutions. The algorithms will be tunned and implemented by professionals in the area of Biomedical Engineering at the Institute of Science and Technology of the Federal University of São Paulo, São José dos Campos Unit (ICT -UNIFESP) and the Institute of Mathematics and Computer Sciences at the University of São Paulo, São Carlos Unit (ICMC-USP). A sample will be selected, retrospectively, by surveying lesions compatible with histological diagnosis within the following categories: (1) oral potentially malignant disorders, (2) squamous cell carcinoma, (3) salivary gland tumors and (4) lymphomas. The deep learning neural networks used in this context will be AlexNet, ResNet, DenseNet, Inception, Xception and MobileNet. Eventually, the best architecture will be selected. The performance of each approach used will be calculated using parameters of accuracy, sensitivity, specificity, F1 score coefficient, ROC curve graphs and confusion matrices. The prognostic significance will be investigated using Kaplan-Meier curves and Cox summit analyses, performing univariate and multivariate analyses of digital, clinical and pathological parameters.
News published in Agência FAPESP Newsletter about the scholarship: