Project description
The main objective of this project is to design and apply techniques based on Deep Learning for the processing of medical images from TAC and abdominal resonance in order to minimize the possible artifacts in the image produced by poor positioning of the patient during the test, his movement during the test or a technical problem during the acquisition of the image, among others.
The medical images will be used for the training of a semantic segmentation network for the diagnosis of potentially cancerous abdominal lesions and it is because of this that the accuracy in the detection is crucial. Starting from a training dataset whose images have artifacts can negatively affect this accuracy.
With the objective of developing light and precise neural networks that allow easy integration in hardware systems, the application of mathematical methods for lightening neural networks such as the minimization of parameters and operations through the reuse of layers and the combination of operations will be studied.
Likewise, the study of the impact of the different configuration parameters of the TAC and resonance machine on the resulting image and the development of an algorithm for processing them that is agnostic and independent of these parameters is another of the objectives of this project.
