| Title: | Hyperbolic metric learning in machine learning algorithms for application in oncology : doctoral dissertation |
|---|
| Authors: | ID Trpin, Alenka (Author)
ID Boshkoska, Biljana Mileva (Mentor) More about this mentor... 
ID Mesti, Tanja (Comentor) |
|---|
| Files: |  DR_2024_Alenka_Trpin.pdf (6,67 MB)
MD5: F2A67E89C7F7995C8D9D4E2A71CF360B
|
|---|
| Language: | English |
|---|
| Work type: | Doctoral dissertation |
|---|
| Typology: | 2.08 - Doctoral Dissertation |
|---|
| Organization: | FIŠ - Faculty of Information Studies in Novo mesto
|
|---|
| Abstract: | Machine learning (ML), a subset of artificial intelligence (AI), enables systems to autonomously learn and adapt without continuous human supervision, leveraging algorithms to process data, identify patterns, and refine performance through experience. This adaptive, selfteaching capability allows ML models to enhance their predictive accuracy and efficiency, making them suitable for dynamic and complex tasks. This dissertation introduces a novel approach to independent and efficient image classification, combining elements from convolutional neural networks (CNNs), hyperbolic geometry, and feature extraction. Unlike existing methods that typically rely on one of these techniques, our integrated approach merges their strengths to achieve superior performance. Additionally, we
developed derivative methods based on the original approach, which enhanced capabilities in embedding data in space. All proposed techniques were empirically evaluated on both image and numerical datasets, consistently demonstrating superior performance when compared to baseline methods. Comparative analysis confirmed that our approach achieves higher classification accuracy than traditional techniques. Given the critical role of accurate and efficient diagnostic tools in oncology, where vast amounts of data from various patient examinations need to be processed, the development of robust algorithms is essential for effective cancer diagnosis and treatment. This dissertation specifically addresses cancer image classification, focusing on the differentiation between benign and malignant lesions – an essential task for early cancer detection and treatment. Empirical results showed that embedding the data in a hyperbolic space, combined with the method for metric learning Large Margin Nearest Neighbours (LMNN) method and the use of Poincaré distance in the k-Nearest Neighbours (kNN) algorithm, yielded comparable or superior results compared to traditional classification techniques. Our findings highlight the
potential of hyperbolic embeddings and metric learning approaches to advance image classification in oncology, offering a promising direction for further research and clinical applications. |
|---|
| Keywords: | cancer images, convolutional neural network, embedding data, hyperbolic geometry, image classification, k-nearest neighbours method |
|---|
| Place of publishing: | Novo mesto |
|---|
| Place of performance: | Novo mesto |
|---|
| Publisher: | A. Trpin |
|---|
| Year of publishing: | 2024 |
|---|
| Year of performance: | 2024 |
|---|
| Number of pages: | XXIII, 157 str. |
|---|
| PID: | 20.500.12556/ReVIS-11273 |
|---|
| COBISS.SI-ID: | 221306115 |
|---|
| UDC: | 004.93:616-006(043.2) |
|---|
| Note: | Na ov.: Doctoral dissertation;
|
|---|
| Publication date in ReVIS: | 08.01.2025 |
|---|
| Views: | 153 |
|---|
| Downloads: | 3 |
|---|
| Metadata: |    |
|---|
|
:
|
Copy citation |
|---|
| | | | Share: |  |
|---|
Hover the mouse pointer over a document title to show the abstract or click
on the title to get all document metadata. |
