Main objective: develop a software system for computer aided and automatic diagnosis of the abdominal tumors based on multiple image modalities, involving both conventional and deep learning techniques.
Secondary objectives:
The project activities are organized in three stages, corresponding to each year.
In the context of the ACADTUM research project, the research team developed and experimented advanced computerized methods for the automated and computer aided diagnosis of the abdominal tumors, based on medical images of various types: ultrasound (US), computer tomography (CT), magnetic resonance images (MRI). Aiming to perform abdominal tumor recognition and segmentation within medical images, representative Convolutional Neural Networks (CNN) based techniques, original CNN architectures, as well as CNN combinations, at classifier and decision level, were considered for this purpose. Original conventional techniques, based on advanced texture analysis methods, were experimented as well, being compared, respectively combined with the deep-learning methods. Important steps have been performed to automatically detect the renal tumors’ evolution stages, respectively the pre-neoplastic states, in the case of liver cancer. The best performing techniques were integrated within the ACADTUM software system, destined for the automatic and computer aided diagnosis of abdominal tumors. The automatic recognition and segmentation methods led to a superior performance when being assessed on CT and MRI images. However, the value of the ultrasound imaging based medical examination techniques should not be ignored, being known that ultrasonography represents a non-invasive, low-cost, safe medical investigation method, suitable for disease evolution monitoring.
Delia Mitrea
Address:
Technical University of Cluj-Napoca
Computer Science department,
Memorandumului st., no. 28, 400 114, Cluj-Napoca
Romania
Office: Baritiu st., no. 26
E-mail:
Delia.Mitrea@cs.utcluj.ro