Here is a list of suggested project ideas for BSc and MSc students at the School of Computer Science, University of Leeds. Students interested in computational biomedicine and materials are encouraged to apply. We focus on the application of deep learning to understand biomaterials, biological tissues, and naturally occurring micro-structures. The projects will be mainly supervised by Dr. Arash Rabbani. Contact: a.rabbani@leeds.ac.uk

1- Deep learning for improving cell segmentation and classification in histology images

In this project we will be using publicly available datasets of histopathology images to develop and benchmark deep learning models for nuclei instance segmentation and classification. The student will explore state-of-the-art architectures such as Vision Transformers (CellViT) and HoVer-Net to segment and classify cell nuclei from H&E stained tissue slides. The result of this study will be a model capable of precise detection and classification of nuclei types (neoplastic, inflammatory, connective, dead, and epithelial) with performance metrics or computational efficiency competing with the existing baselines in the literature.

A related paper:

Hörst, F., et al. (2024). "CellViT: Vision Transformers for precise cell segmentation and classification." Medical Image Analysis, 94, 103143. DOI: 10.1016/j.media.2024.103143

Useful repository:
(Link)

Dataset: PanNuke Dataset (nearly 200,000 annotated nuclei across 19 tissue types with 5 clinically important cell classes)
(Link)

2- Deep learning based cardiac MRI segmentation and pathology classification

In this project, the student will develop deep learning models to segment cardiac structures (left ventricle, right ventricle, and myocardium) from cine MRI images and classify cardiac pathologies. One possible goal for this project is to evaluate the performance and computational efficiency of different CNN-based and transformer-based methods for automated cardiac diagnosis. The student will work on predicting physiological parameters including diastolic volume, ejection fraction, and myocardial mass from the segmented structures.

A related paper:

Bernard, O., et al. (2018). "Deep Learning Techniques for Automatic MRI Cardiac Multi-Structures Segmentation and Diagnosis: Is the Problem Solved?" IEEE Transactions on Medical Imaging, 37(11), 2514-2525. DOI: 10.1109/TMI.2018.2837502

Useful repositories:
(Link)

and:
(Link)

Dataset: ACDC Challenge Dataset (150 cine MRI exams from patients divided into 5 subgroups: normal, previous myocardial infarction, dilated cardiomyopathy, hypertrophic cardiomyopathy, and abnormal right ventricle)
(Link)

3- Conditional Generative Adversarial Networks for materials design with controlled properties

In this project, the student needs to develop a conditional generative adversarial neural network model that is capable of generating pore-scale images of micro-structured materials with precisely controlled physical properties. The generated images should capture realistic micro-structural features while satisfying user-defined constraints on porosity, permeability, or other petrophysical attributes. The outcome of this project will be a controllable data augmentation and materials design framework that addresses the challenge of limited availability of imaging data.

A related paper:

Sadeghkhani, A., et al. (2025). "PCP-GAN: Property-Constrained Pore-scale image reconstruction via conditional Generative Adversarial Networks." arXiv preprint arXiv:2510.19465.

Useful repository:
(Link)

Dataset: Custom dataset as described in the paper (available on request), or publicly available digital datasets from the Digital Rocks Portal
(Link)