A CNN-based methodology for breast cancer diagnosis using thermal images (1910.13757v1)

Abstract: Micro Abstract: A recent study from GLOBOCAN disclosed that during 2018 two million women worldwide had been diagnosed from breast cancer. This study presents a computer-aided diagnosis system based on convolutional neural networks as an alternative diagnosis methodology for breast cancer diagnosis with thermal images. Experimental results showed that lower false-positives and false-negatives classification rates are obtained when data pre-processing and data augmentation techniques are implemented in these thermal images. Background: There are many types of breast cancer screening techniques such as, mammography, magnetic resonance imaging, ultrasound and blood sample tests, which require either, expensive devices or personal qualified. Currently, some countries still lack access to these main screening techniques due to economic, social or cultural issues. The objective of this study is to demonstrate that computer-aided diagnosis(CAD) systems based on convolutional neural networks (CNN) are faster, reliable and robust than other techniques. Methods: We performed a study of the influence of data pre-processing, data augmentation and database size versus a proposed set of CNN models. Furthermore, we developed a CNN hyper-parameters fine-tuning optimization algorithm using a tree parzen estimator. Results: Among the 57 patients database, our CNN models obtained a higher accuracy (92\%) and F1-score (92\%) that outperforms several state-of-the-art architectures such as ResNet50, SeResNet50 and Inception. Also, we demonstrated that a CNN model that implements data-augmentation techniques reach identical performance metrics in comparison with a CNN that uses a database up to 50\% bigger. Conclusion: This study highlights the benefits of data augmentation and CNNs in thermal breast images. Also, it measures the influence of the database size in the performance of CNNs.

• The paper presents a novel CNN-based approach that uses thermal imaging to achieve 92% accuracy and F1-score in breast cancer diagnosis.
• The paper demonstrates that simpler CNN architectures, optimized via Bayesian methods, outperform complex models by reducing bias and overfitting.
• The paper highlights that data augmentation can effectively counter small dataset sizes, reducing necessary patient samples by 50% for reliable performance.

A CNN-based Methodology for Breast Cancer Diagnosis Using Thermal Images: An Analytical Perspective

The paper presents a novel approach for breast cancer diagnosis using thermal images through a Convolutional Neural Network-based (CNN) methodology. This research is motivated by the need for cost-effective, reliable, and non-invasive screening alternatives to traditional techniques such as mammography and MRI, which often require expensive equipment and skilled personnel.

Methodology and Key Findings

The paper leverages thermal imaging's potential, a relatively low-cost, radiation-free alternative, combined with computer-aided diagnosis (CAD) systems that employ CNNs for improved pattern recognition. The research utilizes the DMR-IR database, a widely recognized source comprising thermal images from 57 patients, of which 38 present abnormalities.

Several methodological innovations and experiments are conducted:

1. Baseline and State-of-the-Art Model Comparison: The paper critiques traditional database-splitting techniques, demonstrating their propensity to cause bias and overfitting. By restructuring the database into distinct patient-based training and test sets, the authors achieve improved reliability in their CNN models.
2. Benchmarking and Architecture Simplification: Employing various state-of-the-art CNN architectures, the paper finds that simpler architectures like SeResNet18 outperform more complex ones such as ResNet50 and InceptionResNet V2 in this context, suggesting that the complexity of such architectures might be redundant for the inherent pattern recognition required by the DMR-IR dataset.
3. Hyper-parameters Optimization: Utilizing a Bayesian optimization approach with a Tree Parzen Estimator, the paper fine-tunes the CNN architectures to achieve a 92% accuracy and F1-score, outperforming traditional texture and feature-based methods.
4. Data Augmentation and Database Size: A crucial finding is the effectiveness of data augmentation, which compensates for the limited size of biomedical datasets. The paper reveals that data augmentation techniques can achieve performance metrics equivalent to a significantly larger dataset, reflecting a 50% reduction in necessary patient samples.

Implications and Future Directions

This research holds significant implications for the development of CAD systems in breast cancer screening, particularly in resource-constrained environments. The demonstrated efficacy of CNN-based systems using thermal imaging could enhance accessibility to early diagnosis in areas where traditional screening methods are not feasible due to financial or infrastructural limitations.

Practically, this paper advocates for the application of CNNs in improving diagnostic accuracy and reducing false positives and negatives, which are critical for effective cancer screening. However, the authors also acknowledge the limitations of current datasets and advocate for the creation of larger, high-quality databases to further validate their approach.

Theoretically, the work challenges the trend towards ever more complex CNN architectures by showing that simplicity can sometimes yield better results. This suggests future research directions in machine learning, urging the exploration of tailored, dataset-specific model simplifications and optimizations.

In conclusion, the paper presents compelling evidence for a CNN-driven diagnostic framework utilizing thermal images for breast cancer detection. This innovative approach not only strengthens the potential of non-invasive diagnostic techniques but also paves the way for more cost-effective and scalable solutions across various medical imaging applications. The authors' methodological rigor and empirical findings make a substantial contribution to the ongoing advancement of AI-driven diagnostic tools in healthcare.