Optics and Precision Engineering, Volume. 31, Issue 6, 936(2023)
Bone scintigraphic classification method based on ACGAN and transfer learning
Figures & Tables(22)
Table 1. Number and ratio of training and testing of bone imaging
View tableView in ArticleTable 1. Number and ratio of training and testing of bone imaging
Index Health Metastases Benign changes Train/Test Train/Test Train/Test Quantity 130/32 101/25 43/15 Total 162 126 58 Ratio 46.82% 36.42% 16.76%
Table 2. Amount of bone imaging data after the expansion
View tableView in ArticleTable 2. Amount of bone imaging data after the expansion
Class Health Metastases Benign changes Quantity 1 950 1 515 645
Table 3. Quality assessment of bone imaging
View tableView in ArticleTable 3. Quality assessment of bone imaging
Image type Health Metastases Benign changes ACGAN 0.458 5 0.516 0 0.506 8 MU-ACGAN 0.539 0 0.557 2 0.548 5
Table 4. Classification effects of different data augmentation methods
View tableView in ArticleTable 4. Classification effects of different data augmentation methods
Number Accuracy Precision Recall Specificity F1-socre 1 0.701 3 0.695 3 0.665 7 0.680 1 0.841 6 2 0.727 3 0.730 1 0.687 8 0.708 3 0.845 2 3 0.740 3 0.737 5 0.675 2 0.705 0 0.850 8 4 0.779 2 0.803 2 0.692 8 0.743 9 0.868 8 5 0.805 2 0.866 2 0.728 4 0.791 3 0.882 5
Table 5. Comparison of transfer learning strategies
View tableView in ArticleTable 5. Comparison of transfer learning strategies
Type Accuracy Recall Transfer learning 0.805 2 0.728 4 Dataset5 0.662 3 0.589 9 Training directly 0.597 4 0.527 6
Table 6. Experimental evaluation of multi-scale CNN model ablation
View tableView in ArticleTable 6. Experimental evaluation of multi-scale CNN model ablation
Models Accuracy Recall F1-socre VGG16 0.766 2 0.666 3 0.743 6 DenseNet121 0.805 2 0.755 0 0.767 7 ResNext50 0.766 2 0.696 8 0.736 7 Multi-scale CNN 0.831 2 0.763 9 0.817 8
Table 7. Comparison of the experimental results of different CNN models
View tableView in ArticleTable 7. Comparison of the experimental results of different CNN models
Models Accuracy Precision Recall Specificity F1-socre ResNet34 0.779 2 0.803 2 0.692 8 0.868 8 0.743 9 MobileNetV3 0.779 2 0.785 7 0.745 9 0.875 6 0.765 2 InceptionV3 0.792 2 0.847 7 0.750 6 0.874 2 0.796 2 EfficientNet 0.805 2 0.868 4 0.741 8 0.880 6 0.800 1 Ours 0.831 2 0.879 9 0.763 9 0.899 2 0.817 8
Table 8. Evaluation metrics of different CNN classification models
View tableView in ArticleTable 8. Evaluation metrics of different CNN classification models
Dataset Accuracy Recall Specificity Dataset4 0.831 2 0.763 9 0.899 2 Dataset5 0.857 1 0.804 0 0.912 0
Tools
Get Citation
Copy Citation Text
Hong YU, Renze LUO, Chunmeng CHEN, Xiang TANG, Renquan LUO. Bone scintigraphic classification method based on ACGAN and transfer learning[J]. Optics and Precision Engineering, 2023, 31(6): 936
Paper Information
Category: Information Sciences
Received: Jul. 19, 2022
Accepted: --
Published Online: Apr. 4, 2023
The Author Email: YU Hong (790622472@qq.com)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20