Related Experiment Video
Updated: Aug 13, 2025
03:31
Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications
Published on: December 15, 2023
600
Breast cancer detection: Shallow convolutional neural network against deep convolutional neural networks based
Himanish Shekhar Das1, Akalpita Das2, Anupal Neog3
1Department of Computer Science and Information Technology, Cotton University, Guwahati, India.
Frontiers in Genetics
|January 23, 2023
Summary
This study demonstrates that deep convolutional neural networks significantly improve breast cancer (BC) detection accuracy on mammograms compared to shallow networks. Fine-tuned deep learning models achieved superior performance in identifying malignant tumors, aiding radiologists in diagnosis.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Oncology
Background:
- Breast cancer (BC) is a leading cause of cancer mortality in women globally.
- Accurate diagnosis of breast abnormalities from mammograms is crucial for effective treatment.
- Computer-aided diagnosis (CAD) systems aim to enhance radiologist accuracy.
Purpose of the Study:
- To compare the diagnostic performance of shallow convolutional neural networks (CNNs) against pre-trained deep CNN architectures for breast cancer detection.
- To evaluate the effectiveness of transfer learning in improving BC identification from mammographic images.
Main Methods:
- Mammogram images were pre-processed for automatic breast cancer identification.
- Three shallow CNN architectures were trained on the processed images.
- Transfer learning via fine-tuning was applied to pre-trained deep CNNs (VGG19, ResNet50, MobileNet-v2, Inception-v3, Xception, Inception-ResNet-v2).
Main Results:
- The study achieved accuracies of 80.4% and 89.2% on the CBIS-DDSM dataset.
- Accuracies of 87.8% and 95.1% were recorded on the INbreast dataset.
- Deep network-based approaches with fine-tuning outperformed shallow networks and other state-of-the-art methods.
Conclusions:
- Deep learning models, particularly when fine-tuned, offer superior performance for breast cancer detection in mammography.
- The findings support the integration of advanced deep CNNs into CAD systems to assist radiologists.
- Optimized deep network architectures show significant potential for improving diagnostic accuracy in breast cancer screening.
