Distilling Knowledge From an Ensemble of Vision Transformers for Improved Classification of Breast Ultrasound

被引:10
作者
Zhou, George [1 ]
Mosadegh, Bobak [2 ]
机构
[1] Weill Cornell Med, New York, NY 10021 USA
[2] Weill Cornell Med, Dalio Inst Cardiovasc Imaging, Dept Radiol, New York, NY USA
关键词
Breast ultrasound; Deep learning; Vision transformer; Ensemble learning; Knowledge distillation; NEURAL-NETWORK; CANCER; MAMMOGRAPHY;
D O I
10.1016/j.acra.2023.08.006
中图分类号
R8 [特种医学]; R445 [影像诊断学];
学科分类号
1002 ; 100207 ; 1009 ;
摘要
Rationale and Objectives: To develop a deep learning model for the automated classification of breast ultrasound images as benign or malignant. More specifically, the application of vision transformers, ensemble learning, and knowledge distillation is explored for breast ultrasound classification. Materials and Methods: Single view, B-mode ultrasound images were curated from the publicly available Breast Ultrasound Image (BUSI) dataset, which has categorical ground truth labels (benign vs malignant) assigned by radiologists and malignant cases confirmed by biopsy. The performance of vision transformers (ViT) is compared to convolutional neural networks (CNN), followed by a comparison between supervised, self-supervised, and randomly initialized ViT. Subsequently, the ensemble of 10 independently trained ViT, where the ensemble model is the unweighted average of the output of each individual model is compared to the performance of each ViT alone. Finally, we train a single ViT to emulate the ensembled ViT using knowledge distillation. Results: On this dataset that was trained using five-fold cross validation, ViT outperforms CNN, while self-supervised ViT outperform supervised and randomly initialized ViT. The ensemble model achieves an area under the receiver operating characteristics curve (AuROC) and area under the precision recall curve (AuPRC) of 0.977 and 0.965 on the test set, outperforming the average AuROC and AuPRC of the independently trained ViTs (0.958 +/- 0.05 and 0.931 +/- 0.016). The distilled ViT achieves an AuROC and AuPRC of 0.972 and 0.960. Conclusion: Both transfer learning and ensemble learning can each offer increased performance independently and can be sequentially combined to collectively improve the performance of the final model. Furthermore, a single vision transformer can be trained to match the performance of an ensemble of a set of vision transformers using knowledge distillation.
引用
收藏
页码:104 / 120
页数:17
相关论文
共 50 条
[41]   Deep learning and genetic algorithm-based ensemble model for feature selection and classification of breast ultrasound images [J].
Dar, Mohsin Furkh ;
Ganivada, Avatharam .
IMAGE AND VISION COMPUTING, 2024, 146
[42]   A Multi-Task Learning Framework for Automated Segmentation and Classification of Breast Tumors From Ultrasound Images [J].
Chowdary, Jignesh ;
Yogarajah, Pratheepan ;
Chaurasia, Priyanka ;
Guruviah, Velmathi .
ULTRASONIC IMAGING, 2022, 44 (01) :3-12
[43]   Classification of breast cancer from histopathology images using an ensemble of deep multiscale networks [J].
Karthik, R. ;
Menaka, R. ;
Siddharth, M. V. .
BIOCYBERNETICS AND BIOMEDICAL ENGINEERING, 2022, 42 (03) :963-976
[44]   Classification of Breast Cancer Lesions in Ultrasound Images by Using Attention Layer and Loss Ensemble in Deep Convolutional Neural Networks [J].
Kalafi, Elham Yousef ;
Jodeiri, Ata ;
Setarehdan, Seyed Kamaledin ;
Lin, Ng Wei ;
Rahmat, Kartini ;
Taib, Nur Aishah ;
Ganggayah, Mogana Darshini ;
Dhillon, Sarinder Kaur .
DIAGNOSTICS, 2021, 11 (10)
[45]   Ensemble Deep-Learning-Enabled Clinical Decision Support System for Breast Cancer Diagnosis and Classification on Ultrasound Images [J].
Ragab, Mahmoud ;
Albukhari, Ashwag ;
Alyami, Jaber ;
Mansour, Romany F. .
BIOLOGY-BASEL, 2022, 11 (03)
[46]   Breast tumor classification through learning from noisy labeled ultrasound images [J].
Cao, Zhantao ;
Yang, Guowu ;
Chen, Qin ;
Chen, Xiaolong ;
Lv, Fengmao .
MEDICAL PHYSICS, 2020, 47 (03) :1048-1057
[47]   Breast Cancer Classification from Histopathological Images Based on Improved Inception Model [J].
Li Zhaoxu ;
Song Tao ;
Ge Mengfei ;
Liu Jiaxin ;
Wang Hongwei ;
Wang Jia .
LASER & OPTOELECTRONICS PROGRESS, 2021, 58 (08)
[48]   Classification of Invasive Ductal Carcinoma from histopathology breast cancer images using Stacked Generalized Ensemble [J].
Kumar, Deepika ;
Batra, Usha .
JOURNAL OF INTELLIGENT & FUZZY SYSTEMS, 2021, 40 (03) :4919-4934
[49]   A Simple Ultrasound Based Classification Algorithm Allows Differentiation of Benign from Malignant Breast Lesions by Using Only Quantitative Parameters [J].
Kapetas, Panagiotis ;
Woitek, Ramona ;
Clauser, Paola ;
Bernathova, Maria ;
Pinker, Katja ;
Helbich, Thomas H. ;
Baltzer, Pascal A. .
MOLECULAR IMAGING AND BIOLOGY, 2018, 20 (06) :1053-1060
[50]   Improved breast ultrasound tumor classification using dual-input CNN with GAP-guided attention loss [J].
Zou, Xiao ;
Zhai, Jintao ;
Qian, Shengyou ;
Li, Ang ;
Tian, Feng ;
Cao, Xiaofei ;
Wang, Runmin .
MATHEMATICAL BIOSCIENCES AND ENGINEERING, 2023, 20 (08) :15244-15264