Related Experiment Video
Updated: Jan 7, 2026
04:23
A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images
Published on: April 21, 2023
2.2K
Deep Learning for Tumor Segmentation and Multiclass Classification in Breast Ultrasound Images Using Pretrained
K E ArunKumar1,2, Matthew E Wilson1,2, Nathan E Blake1,2
1School of Agriculture and Food Systems, Davis College of Agriculture and Natural Resources, West Virginia University, Morgantown, WV 26506, USA.
Sensors (Basel, Switzerland)
|December 31, 2025
Summary
This study introduces an artificial intelligence (AI) workflow using deep learning for breast cancer detection in ultrasound images. The AI models achieved high accuracy in segmenting lesions and classifying them as normal, benign, or malignant.
Area of Science:
- * Medical Imaging and Diagnostics
- * Artificial Intelligence in Healthcare
- * Biomedical Engineering
Background:
- * Early breast cancer detection is crucial for effective treatment and improved patient outcomes.
- * Breast ultrasound is a widely utilized imaging modality for lesion assessment by radiologists.
- * Existing imaging technologies require enhancement for automated and accurate lesion identification and classification.
Purpose of the Study:
- * To develop artificial intelligence (AI) tools for segmenting lesions and multiclass classification of breast cancer in ultrasound images.
- * To evaluate the performance of various pretrained deep learning models for segmentation and classification tasks.
- * To enhance AI model performance using frequency-domain feature enhancement techniques.
Main Methods:
- * Development of segmentation models using encoder-decoder architectures (U-Net, U-Net++, DeepLabV3) with pretrained backbones (ResNet18, EfficientNet-B0, MobileNetV2).
- * Integration of a Frequency-Domain Feature Enhancement Module (FEM) into the DeepLabV3 architecture for improved feature representation.
- * Training and evaluation of multiclass classification models (ResNet18, DenseNet121, InceptionV3, MobielNetV3, GoogleNet) on the Breast Ultrasound Images (BUSI) dataset.
- * Hyperparameter optimization using Optuna for selecting optimal model architectures and parameters.
Main Results:
- * The FrequencyAwareDeepLabV3 (FADeepLabV3) model achieved a Dice Similarity Coefficient (DSC) of 0.85 and Intersection over Union (IoU) of 0.72 on the test dataset.
- * Frequency-domain enhancement significantly improved the area under the curve (AUC) from 0.90 to 0.98, indicating enhanced prediction confidence.
- * For classification, InceptionV3 demonstrated the best performance on the test dataset with an F1 score of 0.75 and accuracy of 0.83.
Conclusions:
- * The developed AI workflow effectively automates the segmentation and multiclass classification of breast cancer ultrasound images.
- * The proposed FADeepLabV3 model with frequency-domain enhancement shows significant potential for improving diagnostic accuracy in breast cancer detection.
- * This approach offers a promising tool for enhancing the clinical reliability and efficiency of breast ultrasound interpretation.
