Deep Learning for Medical Image Segmentation (DLTMIS) is the my undergraduate capstone project at the University of Windsor under the supervision of Dr. Sherif Saad.
The purpose of this project is to assist with the use of available deep learning techniques for medical image segmentation, classification and analysis using existing information from literature and available opensource tools and libraries for developing deep learning solutions. This is to be done by creating a publicly available repository/archive of relevant tools, libraries, applications, resources, etc.
The project aims to deliver a number of opensource products available at no cost to the end user. These deliverable products include:
- a toolkit of resources from different sources for non-data science experts
- dataset catalogue containing links to the dataset and description of data in each dataset
- implementation of different deep learning techniques for medical image segmentation
- deployment and testing instructions
- implementation user guides
- final Report
- user friendly UI (optional)
The project requirements are:
- relevant background and knowledge of medical image segmentation, classification and analysis
- relevant background in machine learning and deep learning
- the use of popular data science languages and libraries such as Python (NumPy, SciPy, Pandas, Scikit-Learn, Jupyter and IPython libraries), R and Java
- access to medical image datasets
- use of suitable software and hardware resources (i.e. Git, Codelabs, etc)
We’re not expecting to design or create new algorithms, architectures or techniques. We’re not creating constructing new datasets. The tools and the scripts are for educational purposes only. The outcomes of this project (tools and script) are mainly for educational purposes. The outcomes of this project are not expected to be used for the production environment.
Potential stakeholders include students studying medical image processing/pathology, students studying applied deep learning in medical image processing and developing teams which include software developers, ML engineers, and testing/quality engineers.
The stakeholders are the end-users who are expected to use the system and provide feedback to contribute to defining system requirements and specification.
This project makes the following assumptions:
- datasets are available at no cost
- stakeholders will contribute in defining the system requirements and specifications as needed
- available deep learning tools and techniques are suitable for medical image segmentation, classification and analysis
- opensource software libraries are sufficient for completing the project
- building the relevant backgrounds as described above is feasible for the project timeline
- the timeline is suitable for completing the project on time and while meeting the deliverable expectations
Time and budget are the main constraints in addition to the availability of useful and suitable datasets. This project is limited to the use of opensource software tools.