Automated Breast Cancer screening and detection leveraging Modified-UNet Convolution Neural Network and Capsule Neural Networks with X-Ray Images

About the Project

The aim of this study is to create an automated system for breast cancer screening and detection using X-ray images, specifically mammograms. The study will utilize Modified-UNet and the efficiency of CapsNets to improve the accuracy of image classification, ultimately leading to enhanced patient outcomes. This system’s purpose is to facilitate accessible early screening and detection of cancer for a wide range of patients, particularly at lower-level healthcare facilities. Medical personnel will be able to utilize this system effectively, thereby expanding the reach of screening services and enhancing early detection capabilities.

Start Date: October 2024
End Date: September 2025

Research Team

Dr. Edna Chebet Too, Ph.D. – Senior Lecturer, echebet@chuka.ac.ke, 0722174058 – Principal Investigator
Prof. David Gitonga Mwathi, Ph.D. – Associate Professor, Computer Science, Chuka University – Co-investigator
Prof. Lucy Kawira Gitonga, Ph.D. – Professor, School of Nursing and Public Health, Chuka University – Domain Expert
Dr. Milpah Kwamboka – Oncologist, Chuka General Hospital – Domain Expert
Julius Gachoki Murumia – Medical Departments, Chuka University – Domain Expert
Saif Kinyori – ICT, Chuka University – Research Assistant, AI Expert
Pauline Mwaka – Research Assistant, AI Expert
Mary Wanjiku – Research Assistant, MSC Computer Science Student

Source of Funding

National Cancer Institute of Kenya

5th Floor Social Health Authority Building, Ragathi Road, Upper Hill, Kenya

Funding Organization Website: https://www.ncikenya.go.ke/

Objectives and Goals

To enhance breast cancer screening and early detection through innovative technology, optimizing Modified-UNet and CapsNets for accurate X-Ray image classification and improving patient outcomes.

Project Overview

Breast cancer stands as a primary cancer affecting predominantly women, characterized by a high mortality rate. Timely screening and early detection play pivotal roles in facilitating effective treatment and improving survival rates. However, conventional screening methods like mammography (X-Rays) necessitate manual interpretation by radiologists, a process prone to subjectivity and time consumption. Furthermore, the accuracy of interpreting imaging results heavily relies on the expertise of radiologists, leading to variability in diagnostic precision. Compounded by the limited availability of radiologists in Kenyan healthcare facilities, this results in delayed screening and detection. Consequently, the initiation of cancer management is delayed, contributing to suboptimal outcomes. The aim of this study is to create an automated system for breast cancer screening and detection using X-ray images, specifically mammograms. The study will utilize Modified-UNet and the efficiency of CapsNets to improve the accuracy of image classification, ultimately leading to enhanced patient outcomes. This system’s purpose is to facilitate accessible early screening and detection of cancer for a wide range of patients, particularly at lower-level healthcare facilities. Medical personnel will be able to utilize this system effectively, thereby expanding the reach of screening services and enhancing early detection capabilities.

Expected Outcomes or Impact

Expected Outputs

The primary output of the study is a hybrid of Modified-UNet (M-UNet) and Capsule Neural Network (CapsNet) Model called Capsule-SegNet. Capsule-SegNet model will be tailored specifically for breast cancer X-ray image classification. It will serve as a technological marvel, capable of real-time image analysis, accurate classification, and seamless integration within a user-friendly GUI. This output stands as a testament to the fusion of cutting-edge technology and user accessibility, offering a tangible solution for medical professionals and patients alike.

Impacts on the Society

Streamlined Clinical Workflows: By integrating Capsule-SegNet into the diagnostic process, healthcare professionals can benefit from a more reliable and efficient tool for image analysis. This technology aims to support radiologists by providing a second opinion, reducing the time and cognitive load associated with manual image evaluation, and potentially decreasing diagnostic ambiguities.
Early Detection and Diagnosis: Accurate and efficient breast cancer detection can lead to earlier diagnosis and intervention, potentially improving patient outcomes and reducing healthcare costs associated with late-stage treatments.
Equity and Scalability: The model’s capacity to be scaled up and made accessible can guarantee that its advantages extend to a larger group of people, including marginalized groups who have little access to specialized medical resources.
One doctoral student and one master’s student will be supported.

Key Findings and Progress

Ongoing research.