Background
Breast cancer stands as a prominent global health concern, predominantly impacting women, with 2.3 million new diagnoses and 685,000 deaths reported globally in 2020, according to 2023 World Health Organisation statistics. Despite its prevalence, the disease boasts a curability rate of 99% with early detection and treatment, emphasizing the critical role of screening programs. Mammography emerges as a pivotal diagnostic method, known for its cost-effectiveness, non-invasiveness, and rapid, accurate diagnosis capabilities, as highlighted by Giaquinto et al. (2022) and Hooshmand et al. (2022). However, the intricate nature of mammogram reporting complicates early breast cancer detection. The breast’s complex structure adds a layer of difficulty to manual cancer detection, prone to errors. The process often necessitates multiple revisions to mitigate false detections, incurring both financial and temporal costs. Consequently, automating the reporting process, thereby dehumanizing it, stands as an ideal solution, offering heightened accuracy in breast cancer detection while streamlining the diagnostic workflow.
Purpose and Justification
The purpose of this study is to leverage Computer-Aided Detection (CAD) systems, specifically integrating Convolutional Neural Network (CNN) technology, to enhance the accuracy of breast cancer detection by mitigating false-negative interpretations. CAD systems have become pivotal in the medical field for improving diagnostic precision, and CNN stands out as a renowned algorithm due to its proficiency in image recognition, as established by Li et al. (2023). The proven success of CNN in effectively distinguishing pneumonia, COVID-19, and normal chest X-ray radiographs, as demonstrated by J.L. et al. (2022), highlights its potential application in the domain of breast cancer detection. Despite the strides made with CAD systems, challenges persist in achieving optimal accuracy. The aim of this research is to address the existing limitations and push the boundaries of breast cancer detection through the refinement and application of CNN technology within CAD systems. By building on the success of CNN in discerning diverse medical conditions from imaging data, this study seeks to contribute to the ongoing efforts to overcome challenges in breast cancer detection, ultimately improving patient outcomes. Through rigorous experimentation and validation, the research aims to establish a more effective and reliable framework for the integration of CNN within CAD systems, with the overarching goal of advancing the field of medical imaging and enhancing the accuracy of breast cancer diagnosis.
Problem Domain
The problem domain at the heart of this research revolves around the persistent challenges in breast cancer detection, despite the widespread use of Computer-Aided Detection (CAD) systems designed to minimize false-negative interpretations. Despite the promise of advanced algorithms, including Convolutional Neural Network (CNN) technology, existing implementations have yet to fully unlock their potential in accurately identifying breast cancer, leaving a critical gap in diagnostic precision. This gap is exacerbated by the intricacies of the current mammogram reporting process, which heavily relies on a limited number of experienced radiologists, leading to delayed reporting and the potential misclassification of mammograms (Wong et al., 2023). The human eye’s limitations in visualizing small tumours and discerning subtle shades in mammograms contribute to missed diagnoses, allowing the unchecked spread of cancer cells. To bridge these gaps, our research endeavours to investigate the limitations hindering the success of CAD systems in breast cancer detection. The proposed solution involves the development of an advanced framework that integrates CNN technology, aiming to not only mitigate workload issues for radiologists but also substantially enhance the accuracy of breast cancer diagnosis. This approach holds the potential to significantly improve early detection rates, ultimately contributing to more effective breast cancer management and improved patient outcomes.
Research Questions
The project endeavours to construct a model designed to automate mammogram reporting, enhancing breast cancer detection accuracy. Its purpose is to alleviate the workload on radiologists and contribute to the early detection of breast cancer. In the end of the project, it will aim to answer the following questions:
- To what extent can the integration of Convolutional Neural Network (CNN)-based algorithms enhance the sensitivity and specificity of automated mammogram reporting for breast cancer detection compared to traditional Computer-Aided Detection (CAD) systems?
- What are the primary challenges and limitations inherent in the current implementation of CAD systems for breast cancer detection, and how can the application of CNN technology address and overcome these issues to improve diagnostic accuracy?
- What pre-processing techniques are optimal for preparing digital mammogram datasets before inputting them into CNN algorithms, considering factors such as image quality, noise reduction, and feature extraction?
- How can CNN be effectively implemented in the context of digital mammogram analysis using Python, taking into account the unique characteristics and requirements of mammographic images?
- What specific architectural and training optimizations are crucial for maximizing the performance and accuracy of CNN models in identifying breast cancer from high-resolution mammogram datasets, and how can these optimizations be justified in terms of enhancing model quality and reliability?
