DEEPCAPS : A Hybrid Feedforward And Capsule Network Approach For Robust Cancer Detection

Abstract

In recent years, advancements in deep learning have significantly contributed to the field of medical imaging, particularly in cancer detection. However, existing models often struggle with capturing spatial hierarchies and relationships in complex medical images, which are crucial for accurate diagnosis. In this study, we propose DeepCaps, a novel hybrid model that integrates the strengths of feedforward neural networks and capsule networks to enhance cancer detection. The feedforward component serves as a powerful feature extractor, while the capsule layers provide robust representation and transformation-invariance by capturing the spatial relationships and orientation of features. Our model is designed to address the challenges of traditional convolutional neural networks (CNNs) by introducing capsule layers that encapsulate feature information into vector forms, allowing for dynamic routing between capsules. This architecture not only improves the model's ability to handle variations in cancerous tissue appearance but also enhances interpretability by preserving the instantiation parameters of detected features. We evaluate DeepCaps on several benchmark cancer datasets, demonstrating its superior performance in terms of accuracy, sensitivity, and specificity compared to state-of-the-art methods. Our results indicate that the combination of feedforward neural networks and capsule networks provides a more robust and reliable approach for cancer detection, potentially offering a valuable tool for clinical diagnostics. This research contributes to the growing body of work in deep learning for medical applications, paving the way for more advanced and interpretable models in cancer detection.