Nedim Šišić: Class Probability Distributions of a Neural Network Classifier of Multiple Sclerosis Lesions on Quantitative Susceptibility Mapping

V ponedeljek, 12. junija 2023, bo ob 16.00 uri izvedeno
predavanje v okviru PONEDELJKOVEGA SEMINARJA RAČUNALNIŠTVA IN INFORMATIKE
Oddelkov za Informacijske znanosti in tehnologije UP FAMNIT in UP IAM.

ČAS/PROSTOR: 12. junij 2023 ob 16.00 v FAMNIT-VP2.

————————————-
PREDAVATELJ: Nedim ŠIŠIĆ
————————————-

Nedim Šišić is a 1st year PhD student of Computer Science at UP FAMNIT. He finished his master’s studies in Computer Science at UP FAMNIT, and is now a teaching assistant at various computer science courses.

——————————————————————————————————————————————————————————————–
NASLOV: Class Probability Distributions of a Neural Network Classifier of Multiple Sclerosis Lesions on Quantitative Susceptibility Mapping
——————————————————————————————————————————————————————————————–

POVZETEK:

Multiple sclerosis (MS) is a demyelinating disease, in which the insulating myelin covers of nerve cells are damaged. Remyelination is a repair process that forms new myelin sheaths. MS lesion types are heterogeneous regarding myelin damage and repair and iron content, and may be classified histopathologically. Recently, quantitative susceptibility mapping (QSM) MRI has been shown to classify white matter MS lesions into demyelinated and remyelinated types in vivo. Furthermore, a neural network model demonstrates potential for automatic classification of such lesions into the types based on QSM. However, the model requires further classification accuracy increase and faces the problem of handling lesions unable to be classified by medical experts. The problem represents a significant hurdle for the potential use of the model in clinical scenarios. In this work, the class probability distributions of the model on different remyelinated, demyelinated, and unclassified lesion types are analyzed, to better understand its performance and estimate its behavior on unclassified lesions. The obtained results are important for improving the model and enabling its transition into medical practice.

Seminar bo potekal v angleškem jeziku v predavalnici FAMNIT-VP2.

Vabljeni!