Research and develop innovative procedures and the frontiers of knowledge for advanced medical image processing and its applications in computer-aided diagnosis and computational modeling of human physiological systems.

Today images play an important role in medicine because they allow the internal visualization of the human body in-vivo, and are thus a powerful tool to assist diagnosis. Computer vision, computer graphics and pattern recognition are areas that have evolved rapidly owing to the demand for computational solutions that can contribute new information taken from these images. The area of computer vision encompasses techniques based on physical and mathematical models that perform processing and analysis of images in order to characterize them and thus computationally simulate human vision. Computer graphics comprises methods for data visualization in three dimensions (3D), which realistically simulate structures, and pattern recognition, which classifies information based on a priori knowledge that can be features extracted from data or images. The research activities undertaken in the current INCT-MACC (that finishes in 2015) provided an environment where professionals in computer simulation, computer graphics and medicine collaborated to solve strategic problems in bioengineering. On the one hand, this environment has facilitated the development of tools to meet the requirements in computer graphics and image processing for the projects of the institute. On the other hand, the interaction between the professionals involved has allowed experts in scientific visualization to explore methods in continuum mechanics to develop new algorithms. It is in this dual context that research in this area lies, according to the activities listed below. This line of research aims to develop tools based on these three areas: computer vision, computer graphics and pattern recognition. These tools can distinguish relevant structures in medical images, especially in modalities with millimeter resolution such as intravascular ultrasound, or micrometric resolution such as intravascular optical coherence tomography. In particular these tools allow the geometric reconstruction of medical structures which are relevant for medical diagnosis and for computational modeling of the behavior of physiological systems in normal conditions or altered by disease or medical procedures and their application in surgical planning.

Activities

1. Study and development of mathematical-computational models for image processing of intravascular ultrasound and orthogonal angiograms for 3D reconstruction of coronary vessels and to characterize the phenotype of atheromatous plaques, enabling the identification of fibrous tissue, fibro-lipid, necrotic and calcification through techniques that involve spectral analysis of the radiofrequency (RF) signal.
2. Study and development of mathematical-computational models for the processing of angiographic and magnetic resonance images for 3D reconstruction of cerebral vessels including reconstruction of aneurysms and coronary vessels.
3. Study and development of mathematical-computational models for the analysis of brain aneurysms including characterization of the "parent vessel", the "aneurysmal sac" and "neck" aiming at the automatic definition of morphological indices associated with aneurysm.
4. Development of a mathematical model that represents the different layers of the vascular wall: intima, media and adventitia; further, to differentiate between tissues types and their composition (calcium, lipid or fiber) in intravascular optical coherence tomography (OCT) images, as well as to propose a set of software tools that can distinctly represent different tissues around the vascular wall, present in OCT images and quantify them automatically.
5. Study and development of methods for processing orthopedics image volumes from computerized tomography (CT) scans and magnetic resonance for analysis of biomechanical parameters.
6. Explore methods of interpolation to obtain the three-dimensional reconstruction from a set of transverse sections (images) of an object. Applying methods based on neural networks, statistical learning and wavelet transform to generate descriptors for objects obtained.
7. Development of methods for segmentation via entropy and tracking of medical images with applications in deformable models and optical flow.
8. Development of models for calculating optical flow and image processing using wavelets with applications for tracking.
9. Explore level sets methods and topological derivatives with applications to image registration.
10. Application of mass-spring models and concepts in isogeometric analysis with applications for modeling flexible tissues for medical simulators in virtual reality environments.
11. Application of statistical learning methods (SVM, LDA, PCA, etc.) for image classification, segmentation, retrieval of images and objects in databases and dimensionality reduction.
12. Development of methods for information retrieval and data mining in image banks via methods based on tensorial representation and varieties of learning.
13. Integration of statistical models involving shape and texture, deformable models and learning techniques for the modeling, rendering and animation of faces with a view to applications in craniofacial bioengineering.
14. Development of techniques of additive manufacture to construct three-dimensional physical models of cardiac structures from medical images.
15. Study of the functionality of medical patient-specific oriented devices.

Goals

1. By the end of 2019: Develop a set of methods for quantification of intravascular image structures in optical coherence tomography.
2. By the end of 2019: Develop a set of methods for tissue characterization in intravascular ultrasound images.
3. By the end of 2019: Develop a set of methods for analysis of biomechanical parameters available through a friendly interface.
4. By the end of 2017: Develop methodology for three-dimensional reconstruction, via interpolation, from a set of cross sectional images of an object.
5. By the end of 2017: Develop methodology for the representation of image banks and dimensionality reduction via multilinear algebra (tensors).
6. By the end of 2017: Develop methodology for dimensionality reduction via manifold learning.
7. By the end of 2019: Develop prototype including feature extraction techniques, dimensionality reduction, neural networks, statistical learning, transforms for classification, segmentation and retrieval of images and objects in databases.
8. By the end of 2021: Develop methodology for processing via entropy tracking in medical images, with application to deformable models, optical flow and topological derivative.
9. By the end of 2019: Develop methodology based on mass-spring models and concepts in isogeometric analysis for modeling flexible materials/tissues for medical simulators in virtual reality environments.
10. By the end of 2019: Develop methods for information retrieval and data mining in image banks via methods based on tensorial representation and varieties of learning.
11. By the end of 2021: Develop prototype with integration of statistical learning models, deformable models and rendering for craniofacial applications in bioengineering.
12. By the end of 2019: Improve methods for automatic detection of tuberculosis bacilli (tubercle bacilli) from the development of techniques for image fusion of conventional microscopy.
13. By the end of 2019: Develop an algorithm for differentiation of breast lesions in ultrasound images, from texture characteristics and morphometric parameters, using a novelty classifier.
14. By the end of 2021: Contribute to the achievement of Goal 6 of the "Millennium Development Goals (WHO report, 2013)" through the development of a method for the diagnosis of malaria using microscopy images.
15. By the end of 2019: Construction of three-dimensional physical models of cardiac structures from digital medical images.
16. By the end of 2021: Construction of complex structures from computational imaging combining advanced biocompatible materials and pre-clinic assessment.

Impacts

• Improve quality of diagnosis from medical images.
• Increase efficiency in procurement procedures and the combination of these procedures for decision-making.
• Reduction in operating costs associated with the process of medical images acquisition.
• Make open-source tools for image processing available for the use of the medical community at large.
• Human Resources Development: The research themes in question are being developed with the participation of graduate and undergraduate students from involved institutions.
• Application of additive manufacture in the development of devices and customized medical implants.
• Definition of quantitative scores for assessment of risks and benefits for a given therapy or surgical procedure.