Weakly Supervised Classification for Breast Cancer Grading Using Smartphone-based Microscopy Images

Date

2026-1

Author

Dolkay, Kerem

Metadata

Show full item record

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Item Usage Stats

61
views

0
downloads

Cite This

Breast cancer histopathological grading plays an important role in treatment planning and clinical decision-making. The Nottingham Histological Grading system is the most widely used criterion for grading invasive breast carcinoma. However, man- ual grading is subjective, time-consuming, and highly dependent on expert knowl- edge. While recent deep learning methods have shown promising performance in au- tomated grading using high-resolution whole slide images (WSIs), such approaches rely on expensive imaging infrastructure, which limits their applicability in resource- constrained settings. This thesis develops an automated breast cancer grading ap- proach using multiple low-cost smartphone-based microscopy images. A weakly su- pervised learning scheme is adopted since only patient-level Nottingham grade labels are available. A three-stage pipeline is proposed, consisting of feature extraction us- ing a pretrained pathology foundation model (UNI), patient-level classification via an attention-based weakly supervised model (CLAM), and an ensemble learning strategy that employs pairwise binary classifiers to achieve improved multi-class grading accu- racy. Experimental results demonstrate that the proposed ensemble-based approach consistently outperforms existing baseline methods in terms of various evaluation metrics and shows promising performance. These findings indicate that automated breast cancer grading can be achieved using low-cost smartphone-based microscopy images instead of WSIs, suggesting that computational pathology can be made feasi- ble in resource-limited clinical environments.

Subject Keywords

Breast cancer grading, Computational pathology, Weakly supervised learning, Smartphone-based microscopy, Ensemble learning

URI

https://hdl.handle.net/11511/118739

Collections

Graduate School of Natural and Applied Sciences, Thesis