Radiomics using Deep Learning with High Performance Computing
Imaging has great potential to guide therapy because it can provide a more comprehensive view of the entire tumor and it can be used on an ongoing basis to monitor the development and progression of the disease or its response to therapy.
Machine learning techniques have played an increasingly important role in medical image analysis and now in Radiomics. Recently, deep learning techniques have become the state-of-the-art methods for image processing over traditional machine leaning solutions due to deep learning models capabilities at processing high-dimensional, large-scale raw data.
Radiomics can be divided into distinct processes: I) image acquisition, II) image segmentation, III) 3D rendering, IV) feature extraction and feature qualification and V) bioinformatics analyses, where each of these individual processes poses unique challenges. In this project we propose to use a Convolutional Neural Network (CNN) in a Deep Learning approach to solve the II, IV and V processes. Additionally, CNN feature embedding spaces offers further capabilities such as allowing analogic reasoning through vector operations.
This new approach can provide new information about the image features extraction, image segmentation, feature qualification and bioinformatics analysis. As a requirement, deep learning, needs huge quantity of data to train its models. Successful object image classification using deep learning have been using (10^5 to 10^6 images). We are interested in the National Lung Screening Trial data base due to according to World Health Organization, lung cancer is the most common cancer in the world and this data base has a lot information ( 11.3 TB DICOM) with 21,082,502 images or 73,118 studies.
• Automatic region/ volume of interest detection and segmentation of CT lung images using a deep learning technique.
• Feature qualification in multiples image scales.
• Feature extraction in multiples image scales.
• Bioinformatics analysis and phenotype classification
• Semantic featured vector descriptor
• Propose a Deep learning multimodal for CT lung images
• Obtain a general information and try to compute a transfer learning
Dario Garcia-Gasulla Barcelona Supercomputing Center
Ulises Cortés Barcelona Supercomputing Center