Here, is a list of suggested project ideas for BSc and MSc students at the school of computing, University of Leeds. Students who are interested in application of data-driven methods and deep learning for 2D and 3D image analysis are encouraged. We are keen to create innovative solutions to study micro-structed materials and tissues. The projects will be mainly supervised by Dr. Arash Rabbani.

1- Application of deep learning for characterization of porous material

In this project we will be using the available datasets of porous materials to make deep learning models that can predict physical properties of porous material by analyzing their 2D or 3D images. The result of this study will be a regression AI model that is capable of predicting such properties with a higher accuracy than the common practice in the literature by incorporating innovative and cutting-edge neural network architectures. It is recommended to develop the machine learning model in Python, using Tensorflow and have the repository published on GitHub to increase the impact of the proposed research project.

A related paper with GitHub repository:
(Link)

2- Image resolution enhancement for histological images using attention u-net

In this project, student will use innovative deep learning models to increase the resolution of the histological images and evaluate the performance of their work by comparing the results to out-of-the-sample datasets. The ultimate goal is to check if this resolution improvement can make the automated cell-counting processes or classifications more accurate. It is highly recommended to develop the machine learning model in Python language, using Tensorflow and release the repository on GitHub to increase the impact and visibility of this research project.

A related paper with GitHub repository:
(Link)

3- Supervised machine learning for segmentation of histological slides based on a single image

In this project student is required to explore different methods of data augmentation and compare their performance when the labelled data is scarce. These methods should be tested for different architectures of machine learning models and be repeated for different datasets to see which one is more successful in prediction of out-of-the-sample cases. It is recommended to develop the machine learning model in Python, using Tensorflow and have the repository published on GitHub to increase the impact of this research project.

A related paper with GitHub repository:
(Link)

4- 3D Reconstruction of porous materials based on 2D images using conditional adversarial neural networks

In this project, student needs to develop a generative adversarial neural network model that is capable of generating 3D binary porous material images. The generated images should look reasonably realistic and have certain properties. The outcome of this project will be a controlled data augmentation technique that has a space-filling design. A suggestion is to create the machine learning model using Python language and via Tensorflow. Also, to help to amplify the reach and visibility of the research project it is suggested to make the repository available on GitHub.

A related paper with GitHub repository:
(Link)

5- Using Bayesian optimization to reconstruct 3D structures of porous polymer membranes

In this project, student will use statistical generative methods to create stochastic realizations of a porous structure and using Bayesian optimization, the generator function will be guided to produce structures that are visually similar to the real samples that are created in lab. It is recommended to develop the computational code in Python language and have the repository published on GitHub to increase the impact of this research project.

A related paper with GitHub repository:
(Link)