Integrating Explainable AI with Infrared Imaging and Deep Learning for Breast Cancer Detection | IEEE Conference Publication | IEEE Xplore

Integrating Explainable AI with Infrared Imaging and Deep Learning for Breast Cancer Detection


Abstract:

The interplay between medical imaging and artificial intelligence has set the stage for groundbreaking advancements in early disease detection. In the critical domain of ...Show More

Abstract:

The interplay between medical imaging and artificial intelligence has set the stage for groundbreaking advancements in early disease detection. In the critical domain of breast cancer diagnosis, accurate, swift, and interpretable models are undeniably paramount. This study presents a pioneering approach to breast cancer detection using infrared breast imagery underpinned by the principles of explainable AI. Our methodology’s core lies in a novel Gaussian pyramid-driven denoising autoencoder meticulously tailored for infrared breast images. This sophisticated denoising technique enhances the quality of input infrared breast images. It paves the way for more accurate feature extraction, a cornerstone in the diagnostic process. Capitalizing on this refined input, our research introduces an ensemble classifier, blending the deep learning capabilities of DenseNet201 with the feature-rich outputs of the auto-encoder. This synergistic amalgamation sets a new benchmark in precision for breast cancer detection in infrared images. However, accuracy without understanding remains an incomplete victory. Addressing this, our study integrates an attention-guided Grad-CAM mechanism, shedding light on the model’s decision-making process. This tool accentuates pivotal regions in imagery through saliency-driven heatmaps, offering clinicians an intuitive window into the AI’s verdict. Our research not only pushes the boundaries of precision in breast cancer detection but also champions the cause of transparency and understanding in AI-driven diagnostics, ensuring that technology and human expertise move forward hand in hand.
Date of Conference: 13-15 December 2023
Date Added to IEEE Xplore: 19 February 2024
ISBN Information:
Conference Location: Raipur, India

References

References is not available for this document.