Author(s):
1. Tijana Geroski, University of Kragujevac, Faculty of Engineering, Serbia
2. Ognjen Pavić, Institute for Information Technologies Kragujevac, Serbia
3. Lazar Dašić, Institute for Information Technologies Kragujevac, Serbia
4. Marina Petrović, University Clinical Centre Kragujevac, Serbia
5. Dragan Milovanović, University Clinical Centre Kragujevac, Serbia
6. Nenad Filipović, Univerzitet u Kragujevcu, Serbia

Abstract:
Machine learning has the ability to discover significant and hidden relationships in a data set and finds great application in clinical diagnosis, treatment and prediction of disease development. Medical images from magnetic resonance imaging (MRI), computed tomography (CT) or X-ray are most often used as input data. The standard medical procedure, which uses manual annotation by an expert doctor, showed high variability and poor reproducibility. This paper shows several examples of the application of deep learning, as a subfield of machine learning, in automating the process of analyzing medical images, shortening the time for diagnosis, as well as ensuring high accuracy and reproducibility of results. In the field of cardiovascular diseases, as part of the SILICOFCM project (https://silicofcm.eu/), extraction of characteristic geometric parameters of the heart was performed in ultrasound images for the purposes of early detection of cardiomyopathy. In the field of neurosurgery, an example of the application of deep learning is the segmentation and classification of the level and side of disc herniation based on MRI images, using convolutional neural networks U-net, AlexNet, ResNet5, VGG16 etc. In the field of pulmonology, deep learning is used for the needs of detection of lung diseases on X-ray images, which is realized within the SoftLungX project (http://softlungx.bioirc.ac.rs/).

Key words:
deep learning, medical images, convolutional neural networks

Date of abstract submission:
15.08.2023.

Conference:
Contemporary Materials 2023 - Savremeni materijali

Applied paper from author