Lung cancer is the leading cause of cancer-related deaths in the world. Therapy decision of lung cancer must consider several factors besides tumor staging, like histological subtype and genetic mutations. Radiomics can aid diagnosis, prognosis, and therapy decision of lung cancer by associating quantitative features extracted from medical images with patient clinical information (staging, histology, genomics, and others). One quantitative characterization approach that has been shown to be very promising for medical imaging is based on deep machine learning, in which convolutional neural networks perform automatic extraction of features, and those can provide a new level of representation to the lesions on imaging exam. Therefore, the goal of this project is to investigate deep learning techniques to be applied on computed tomography (CT) exams for the development of quantitative image biomarkers and prediction models based on radiomics for the diagnosis, prognosis, and therapy decision of lung cancer. The project obtained approval from the institutional research board to be developed. A cohort of lung tumors with histology confirmed by biopsy or surgical resection was retrospectively developed. Some convolutional network models will be investigated, like 3D-ResNet e 3D-DenseNet, applied to the CT exams, and associated to patient clinical data. Therefore, deep learning-based radiomics can present great potential on developing quantitative image biomarkers and prediction models to aid the diagnosis, prognosis, and therapy decision of lung cancer.
News published in Agência FAPESP Newsletter about the scholarship: