Attention UNet 모델을 활용한 유방 초음파 영상 분할: IoU 기반 학습 및 AUC, MSE 평가
Breast Ultrasound Image Segmentation Using Attention UNet: IoU-Based Training and Evaluation with AUC and MSE Metrics
1
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
This study developed a deep learning model to effectively segment lesions in breast ultrasound images and evaluated its performance from various perspectives. The model enhanced segmentation performance by incorporating an Attention Mechanism into the existing UNet model, which emphasizes important features in the images while suppressing irrelevant information. A total of 780 breast ultrasound images collected from over 600 female patients were used as training data, and data augmentation techniques were employed to improve the model's generalization performance. The model was trained for 322 epochs and evaluated based on various performance metrics. The main performance metrics achieved were an Intersection over Union (IoU) of 0.826, an Area Under the Curve (AUC) of 0.95, and a Mean Squared Error (MSE) of 0.0357, indicating that the model demonstrated excellent performance in both lesion segmentation and differentiation between benign and malignant findings. Additionally, the precision (93.2%), recall (92.17%), and F1 score (0.9268) further confirmed that the model maintained high accuracy and consistency in detecting and segmenting lesions. However, the performance on the vali- dation data decreased, with an IoU of 0.5695, suggesting a potential risk of overfitting and indicating limitations in the model's generalization capability. Further improvements in performance are expected through additional validation using diverse datasets and optimization of the model structure. The proposed model holds significant potential as a tool to assist clinicians in diagnostic settings and may play a crucial role in the early detection of breast cancer.
Abstract
This study developed a deep learning model to effectively segment lesions in breast ultrasound images and evaluated its performance from various perspectives. The model enhanced segmentation performance by incorporating an Attention Mechanism into the existing UNet model, which emphasizes important features in the images while suppressing irrelevant information. A total of 780 breast ultrasound images collected from over 600 female patients were used as training data, and data augmentation techniques were employed to improve the model's generalization performance. The model was trained for 322 epochs and evaluated based on various performance metrics. The main performance metrics achieved were an Intersection over Union (IoU) of 0.826, an Area Under the Curve (AUC) of 0.95, and a Mean Squared Error (MSE) of 0.0357, indicating that the model demonstrated excellent performance in both lesion segmentation and differentiation between benign and malignant findings. Additionally, the precision (93.2%), recall (92.17%), and F1 score (0.9268) further confirmed that the model maintained high accuracy and consistency in detecting and segmenting lesions. However, the performance on the vali- dation data decreased, with an IoU of 0.5695, suggesting a potential risk of overfitting and indicating limitations in the model's generalization capability. Further improvements in performance are expected through additional validation using diverse datasets and optimization of the model structure. The proposed model holds significant potential as a tool to assist clinicians in diagnostic settings and may play a crucial role in the early detection of breast cancer.
참고문헌 (19)
- Baek JW, "The utility and limitations of breast ultrasound as a screening tool in breast cancer detecti on", master, pp.2012
- Yoon GY, "Sonographic features that can be used to differentiate between small triple-negative breast cancer and fibroadenoma", Ultrasonography, 37(2), pp.149~156 (2018)[DOI]
- Falk T, "U-Net : Deep learning for cell counting, detection, and morphometry", Nat Methods, 16, pp.67~70 (2019)[DOI]
- Ronneberger O, "U-Net: Convolutional networks for biomedical image segmentation", International Conference on Medical Image Computing and Computer-Assisted Intervention, pp.234~241 (2015)
- Oktay O, "Medical Imaging with Deep Learning (MIDL)" (2018)
- Al Qurri A, "Improved UNet with attention for medical image segmentation", Sensors(Basel), 23(20), pp.8589~ (2023)[DOI]
- Tang Y, "Attention-guided U-net for medical image segmentation", IEEE Access, 8, pp.16767~16777 (2020)[DOI]
- Flores WG, "U-Net and SegNet performances on lesion segmentation of breast ultrasonography images", Res Biomed Eng, 80, pp.14~23 (2017)[DOI]
- Al-Dhabyani W, "Dataset of breast ultrasound images", Data Brief, 28, pp.104863~ (2020)[DOI]
- Goodfellow I, "Deep learning" (2016), MIT Press
- Bishop CM, "Pattern recognition and machine learning" (2006), Springer
- Long J, "Fully convolutional networks for semantic segmentation", Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) (2015)
- Milletari F, "V-Net: Fully convolutional neural networks for volumetric medical image segmentation", Proceedings of the International Conference on 3D Vision (3DV) (2016)
- Powers DM, "Evaluation: From precision, recall and F-measure to ROC, informedness, markedness and correlation", Int J Mach Learn Technol (2011)[DOI]
- Powers DM, "Evaluation: From precision, recall and F-measure to ROC, informedness, markedness and correlation", Int J Mach Learn Technol (2011)[DOI]
- Sokolova M, "A systematic analysis of performance measures for classification tasks", Inf Process Manag, 45(4), pp.427~437 (2009)[DOI]
- Fawcett T, "An introduction to ROC analysis", Pattern Recognit Lett, 27(8), pp.861~874 (2006)[DOI]
- Goodfellow I, "Deep learning" (2016), MIT Press
- Flores WG, "U-Net and SegNet performances on lesion segmentation of breast ultrasonography images", Res Biomed Eng, 80, pp.14~23 (2017)[DOI]