An Automated In-Depth Feature Learning Algorithm for Breast Abnormality Prognosis and Robust Characterization from
Tariq Mahmood1,2, Jianqiang Li1,3, Yan Pei4
1The School of Software Engineering, Beijing University of Technology, Beijing 100024, China.
Biology
|September 28, 2021
Summary
This study introduces a deep learning model for mammography, achieving 97.8% accuracy in identifying breast cancer. This advanced approach aids in reducing misdiagnoses and improving patient survival rates.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Oncology
Background:
- Accurate mammography interpretation is crucial for early breast cancer detection and treatment.
- Misinterpretations can lead to unnecessary biopsies and reduced patient survival.
- Deep learning offers a promising alternative to traditional feature-based methods for improved diagnostic accuracy.
Purpose of the Study:
- To develop and evaluate a deep learning model for enhanced breast cancer mass and calcification detection in mammography.
- To compare the performance of various deep convolutional neural networks (DCNNs) and a hybrid ConvNet+SVM model.
- To improve the accuracy of mammogram interpretation and reduce diagnostic errors.
Main Methods:
- Utilized a convolutional neural network (ConvNet)-based deep learning approach for feature extraction.
- Implemented five end-to-end pre-trained and fine-tuned DCNNs (VGGNet, GoogLeNet, MobileNet, ResNet, DenseNet).
- Trained a Support Vector Machine (SVM) algorithm using features extracted by ConvNet for a hybrid model, incorporating data cleaning, preprocessing, and augmentation.
Main Results:
- The ConvNet+SVM deep learning model achieved a high discriminative training accuracy of 97.7% and validating accuracy of 97.8%.
- Compared to other DCNNs, the ConvNet+SVM model significantly outperformed VGGNet16 (90.2%), VGGNet19 (93.5%), GoogLeNet (63.4%), MobileNetV2 (82.9%), ResNet50 (75.1%), and DenseNet121 (72.9%).
- The models were evaluated on three mammography datasets, demonstrating remarkable results.
Conclusions:
- The proposed ConvNet+SVM deep learning model shows significant improvement and validation for pathological practices.
- This approach can reduce the burden on pathologists in predicting clinical outcomes from mammography images.
- The study highlights the potential of deep learning in enhancing the accuracy and efficiency of breast cancer diagnosis.
