The UEF Cancer AI team launched a challenge for estimating breast density from mammograms. Addressing limitations in existing tools, the challenge seeks innovative deep-learning architectures. Participants work with a dataset of 569 mammogram images, aiming to enhance detection accuracy. This initiative aims to contribute to early breast cancer detection and prevention.
The UEF Cancer AI research team, spearheaded by esteemed researcher Dr. Hamid Behravan and generously supported by the Finnish Innovation Fund Sitra, has launched a pioneering endeavor in the realm of medical technology. Their latest initiative revolves around a novel deep-learning challenge aimed at revolutionizing the estimation of breast density from mammograms. This innovative pursuit holds immense promise as high breast tissue density is recognized as a significant risk factor for breast cancer.
Breast density stands for the ratio of dense tissue to fatty tissue within the breast. Studies have indicated a higher susceptibility to breast cancer among women with denser breasts compared to those with predominantly fatty breast tissue. Shockingly, women with very dense breasts face a 4 to 5-fold increased risk of breast cancer when juxtaposed with counterparts having fatty breasts.
Current computer-aided design tools intended for estimating breast density percentages from mammograms possess inherent limitations. These tools often grapple with constraints like their applicability to specific mammogram views, inadequacies in delineating the pectoral muscle comprehensively, and suboptimal performance in the face of diverse data types. Moreover, these tools necessitate intervention from an experienced radiologist to fine-tune the segmentation threshold for identifying dense tissue within the breast area.
The crux of the challenge lies in devising a cutting-edge deep learning architecture that transcends these limitations and autonomously calculates area-based breast percentage density from mammograms. Participants are encouraged to explore a spectrum of methodologies encompassing regression and segmentation techniques.
The training dataset for this challenge comprises 569 mammogram images, while the testing phase involves 149 separate images. To facilitate the participants, a baseline segmentation approach’s source code is accessible via the provided Github repository. Entrants are encouraged not only to utilize this model but also to augment and optimize it for the task of density estimation within the challenge.
Participation in this groundbreaking challenge extends an invitation to contribute meaningfully towards a solution that could potentially revolutionize the early detection and prevention of breast cancer. Entrants will have the opportunity to apply their prowess in deep learning and image analysis to address a tangible real-world issue with profound implications for public health.
The challenge marks a significant step in advancing breast cancer detection through deep learning. Participants’ contributions pave the way for enhanced accuracy in estimating breast density from mammograms. The collaborative efforts aim to positively impact early breast cancer detection and prevention.