| Title |
Evaluation of dynamic contrast in prostate mri for cancerous tissue identification / |
| Translation of Title |
Prostatos vėžinių židinių identifikavimas paremtas MRT dinaminio kontrasto vaizdų analize. |
| Authors |
Surkant, Roman |
| Full Text |
|
| Pages |
29 |
| Keywords [eng] |
dynamic contrast, MRI, segmentation, classification, functional data analysis, support vector machine |
| Abstract [eng] |
Prostate cancer is one of the leading causes of cancer death worldwide and early diagnosis and treatment is critical. Cancer evaluation is done by using different imaging sequences of MRI, each having own acquisition methods and purpose. Dynamic contrast enhancement, one of such imaging types, is useful for detecting tumors due to higher vascular permeability and density. Contrast images indicate which areas of patient's body have higher concentration of contrast agent and how that concentration is changing over time. The pattern of such changes is evaluated by segmenting the prostate using SLIC algorithm applied on a Temporal Variation Matrix (TVM), a novel approach indicating high signal variation zones, and constructing time-signal intensity curves from segmented regions. Such discrete signal curves are then transformed into continuous form by applying B-spline functional data smoothing to create functional curves. Two classification approaches - functional (K-Nearest Neighbors and Nearest Centroid) and machine learning (Support Vector Machine) - are used to model signal curve behavior for malignant tissue identification. K-Nearest Neighbor resulted in highest accuracy of 95.67% and highest recall of 75%, but lowest precision, while one of SVM models achieved precision of 93.75% at the cost of the lowest recall - 9.00%. |
| Dissertation Institution |
Vilniaus universitetas. |
| Type |
Master thesis |
| Language |
English |
| Publication date |
2022 |
