FusedMammoNet: Ensemble of diverse models for multi-class mammogram analysis

Cmputer Vision

Breast cancer remains a significant global health challenge, necessitating accurate and efficient detection methods to improve patient outcomes. Mammography serves as a cornerstone for early diagnosis, yet the interpretation of mammograms can be prone to errors, leading to both false positives and missed diagnoses. In response to this critical issue, this study focuses on harnessing the power of Convolutional Neural Networks (CNNs) for the automated detection of breast cancer in mammographic images. The research investigates a diverse range of deep learning techniques, including popular network architectures such as VGG19, ResNet152, InceptionV3, DenseNet121, MobileNetV2, and EfficientNetB0. Various factors crucial to model performance are explored, such as class weighting strategies, input image dimensions, preprocessing methodologies, transfer learning approaches, dropout rates, and the impact of different mammogram projections. Through a systematic and comprehensive analysis, this project aims to evaluate the effectiveness and efficiency of these deep learning methodologies in the context of breast cancer detection. By employing a divide-and-conquer approach, the study seeks to gain valuable insights into selecting the most suitable techniques for enhancing detection accuracy while minimizing the need for extensive trial and error experimentation. The ultimate goal of this research is to advance automated breast cancer screening by optimizing deep learning models for mammogram analysis. By understanding the nuances of various parameters and their impact on model performance, this study aims to contribute to improved diagnostic accuracy and ultimately enhance patient care in the realm of breast cancer detection. The proposed FusedMammoNet model achieved a test accuracy of 96%, recorded highest AUC-ROC ranged from 0.98-1.00 and both precision and recall ranging from 93% to 94%.

This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.