1. Computer vision tools for the automatic evaluation of collagen VI deficiencies
- Author
-
Ramírez i Márquez, Alex, Universitat Politècnica de Catalunya. Departament de Teoria del Senyal i Comunicacions, Vilaplana Besler, Verónica, and Porta Pleite, Josep Maria
- Subjects
Deep learning (Machine learning) ,Computer Vision ,Convolutional Neural Networks ,Visió per ordinador ,Collagen VI ,Computer-Aided Diagnosis ,Neural networks (Computer science) ,Neuromuscular diseases ,Deep Learning ,Xarxes neuronals (Informàtica) ,Informàtica::Intel·ligència artificial [Àrees temàtiques de la UPC] ,Collagen ,Col·lagen ,Aprenentatge profund - Abstract
Deficiencies in the structure of collagen VI are a common cause of neuromuscular diseases. Such diseases typically require assisted ventilation and result in a severely reduced life expectancy. Collagen VI structural defects are related to mutations of three main genes. Currently the CRISPR technology offers a possibility to correct the wrong genes. However, the regulatory agencies would not approve any treatment without an objective methodology to evaluate its effectiveness. This project aims at providing a computer vision solution to evaluate the state of patients with collagen VI deficiencies. The idea is to provide objective metrics of the patient state from images of muscular tissue obtained with a confocal microscope. Currently some tools are available to this end, but only for low resolution 2D images. This project proposes to extend this previous work to the analysis of high-resolution 3D stacks of images. The project involves the development of classical computer vision tools to derive relevant features from the stacks of images and the use of classification tools to generate an overall evaluation of each patient. This analysis will be complemented with the development of a solution based on the use of a convolutional neural network. To this end, data augmentation techniques will be of primary importance since collagen VI-related problems are rare diseases and, thus, there is a severe lack of training data.
- Published
- 2022