Enhanced Detection of Cancer Lesions Using Convolutional Neural Networks and Feature Fusion

Breast cancer remains a significant global health concern, with early detection crucial for effective treatment and improved patient outcomes. Traditional methods of detecting breast cancer lesions, such as mammography and ultrasound, often rely on subjective interpretation and may lack sensitivity. Convolutional Neural Networks (CNNs) have shown promise in medical imaging analysis due to their ability to automatically extract features from images and classify abnormalities. However, improving the detection accuracy of breast cancer lesions using CNNs remains a challenge. Existing CNN-based approaches for breast cancer lesion detection may suffer from limited sensitivity and specificity, leading to missed diagnoses or false positives. Additionally, extracting discriminative features from medical images with varying resolutions and noise levels presents a significant challenge. In this study, we propose an enhanced detection framework for breast cancer lesions using CNNs and feature fusion. Our method incorporates multiple CNN architectures, including DenseNet, ResNet, and Inception, to capture diverse image features. Furthermore, we employ feature fusion techniques to integrate complementary information from different CNN models. By combining features at multiple levels, our approach aims to improve the robustness and discriminative power of the detection model. Experimental results on a large dataset of breast cancer images demonstrate the effectiveness of our proposed method. The proposed framework achieves a sensitivity of 0.95 and a specificity of 0.92, outperforming state-of-the-art methods by a significant margin. Moreover, the proposed method exhibits an area under the receiver operating characteristic curve (AUC) of 0.97, indicating its superior discriminative ability in distinguishing between malignant and benign lesions. The computational efficiency of the proposed approach is also shows, with an average inference time of 0.2 seconds per image.