BREAST TUMOR DIAGNOSIS USING CLASS ACTIVATION MAP BASED CONVOLUTIONAL NEURAL NETWORK

Wang, Zikui

Permalink

https://hdl.handle.net/2142/124797 Copy

Description

Title

BREAST TUMOR DIAGNOSIS USING CLASS ACTIVATION MAP BASED CONVOLUTIONAL NEURAL NETWORK

Author(s)

Wang, Zikui

Issue Date

2023-05-01

Keyword(s)

Breast cancer; Ultrasound image; Deep learning; Convolutional neural networks

Abstract

Breast cancer remains the second leading cause of cancer death among women. Many researchers have developed different deep learning methods, such as Convolutional Neural Networks (CNNs), to diagnose breast cancer. Class-activation-map (CAM) is a technique used to visualize the concentration of deep learning models. While most researchers use CAM to validate the effectiveness of their methods, our CAM-loss-based CNN shows significant improvement in malignancy classification by actively incorporating CAM in the loss function. We demonstrate that our method, compared with the baseline model (ResNet-18), not only effectively improves localization accuracy but also outperforms in terms of the AUC (area under the receiver operating curve), ACC (accuracy), sensitivity, and precision.

Type of Resource

text

Language

eng

Owning Collections