کنفرانس مهندسی برق ایران

صفحه اصلی / سی و چهارمین کنفرانس بین المللی مهندسی برق

Segmentation and Classification of Breast Tumors in Ultrasound Images using Adaptive Saliency of DeepLabV3+ and ResNet34 Networks

نویسندگان :

Seyede Reyhane Khorashadizade¹ Seyed Hamid Khatami²

1- دانشگاه بیرجند 2- دانشگاه بیرجند

کلمات کلیدی :

Deep learning،Adaptive Spatial-Frequency Saliency،Breast ultrasound images،Convolutional Neural Networks (CNN)،Segmentation،Classification

چکیده :

Breast cancer remains one of the most prevalent and complex cancers, where early detection is crucial yet often challenging. This study proposes a highly accurate deep learning–based model for automatic segmentation and classification of breast cancer tumors in ultrasound images. The model employs convolutional neural networks utilizing the BUSI dataset, integrating the DeepLabV3Plus architecture with a MiT-B5 encoder for segmentation and a ResNet34 encoder for classification. Additionally, an adaptive spatial frequency saliency weighting technique is applied during both preprocessing and training to enhance performance. The proposed model achieved an accuracy of 99.49%, a precision of 99.16%, and a Dice coefficient of 94.77% in the segmentation task, outperforming conventional approaches. These results demonstrate the model’s strong potential to assist clinicians by reducing human error and improving the accuracy of early breast cancer detection.