Subject

Breast cancer remains the most common cancer worldwide, with over two million new cases annually. The radiological evaluation of a mammogram (known as the gold standard in breast screening) is based on the presence of masses, asymmetries and/or presence of MCs, where the latter stand out as a key element. However, characterizing MCs is challenging due to mammography's limitations such as resolution constraints, low contrast, and tissue superposition. Our research typically focuses on analysing MCs using a high-resolution 3D micro-CT scanner. Thus far, we have always assessed MCs individual properties, not cluster properties alike they do in mammography ( well-known to be highly important in cancer diagnosis). We assess individual MCs instead of clusters for two main reasons: (a) we have demonstrated in previous work that there is important diagnostic information in individual MCs found in malignant vs benign lesions, (b) there is no proper cluster properties (alike in mammography) sustained in our micro-CT scans as the tissue is removed out of the breast and hence distorted. However, recent scans at 16um resolution have shown potential in preserving at least a close-proximity cluster information of breast MCs, enabling the need for further exploration (under certain assumptions that need to be raised).

Kind of work

Objective: To explore if deep learning (DL) models trained on MCs clusters can outperform models based on individual MCs. At a larger scope, this thesis aims to assess if incorporating high-resolution and cluster information enhances breast cancer diagnostic accuracy.

Literature review (ETOC: 2 months) – understand the medical problem/challenge. Review current methods in 2D mammography and 3D tomosynthesis, particularly deep convolutional neural networks and transformers, for detecting and classifying MC clusters.

Dataset familiarization (ETOC: 1 months): Work with high-resolution 3D micro-CT scans of complete paraffin blocks containing breast tissue with MCs, familiarize with the dataset's specific characteristics and potential challenges (alike cluster radius, number of clusters to consider per biopsy block, very low number of training images etc).

Implement DL models for MCs cluster and individual classification (ETOC: 6 months): Conduct thorough experiments to compare the effectiveness of cluster-based classification vs individual MC classification on micro-CT scans. Perform optimization steps.

Thesis writing (ETOC: 1 month).

Framework of the Thesis

[1] Brahimetaj, R., Willekens, I., Massart, A.?et al.?Improved automated early detection of breast cancer based on high resolution 3D micro-CT microcalcification images.?BMC Cancer?22, 162 (2022). https://doi.org/10.1186/s12885-021-09133-4

[2] Pesapane, F., Trentin, C., Ferrari, F.?et al.?Deep learning performance for detection and classification of microcalcifications on mammography.?Eur Radiol Exp?7, 69 (2023). https://doi.org/10.1186/s41747-023-00384-3

[3] Alzubaidi, L., Zhang, J., Humaidi, A.J.?et al.?Review of deep learning: concepts, CNN architectures, challenges, applications, future directions.?J Big Data?8, 53 (2021). https://doi.org/10.1186/s40537-021-00444-8

[4] Ayana, Gelan, Kokeb Dese, Yisak Dereje, Yonas Kebede, Hika Barki, Dechassa Amdissa, Nahimiya Husen, Fikadu Mulugeta, Bontu Habtamu, and Se-Woon Choe. 2023. "Vision-Transformer-Based Transfer Learning for Mammogram Classification"?Diagnostics?13, no. 2: 178. https://doi.org/10.3390/diagnostics13020178

Expected Student Profile

Following an MSc in a field related to one or more of the following: Computer Science, Biomedical Engineering, Applied Computer Science - Digital Health.

Strong programming skills (Python).

Experience with image processing and DL.

Interest/Motivation in developing state-of-the-art DL methods and conduct experiments.

Ability to write scientific reports and communicate research results at conferences in English.