| dc.contributor | 圖檔所 | |
| dc.creator (作者) | 羅崇銘 | |
| dc.creator (作者) | Lo, Chung-Ming;Lai, Kuo-Lung | |
| dc.date (日期) | 2024-10 | |
| dc.date.accessioned | 24-Feb-2025 15:55:36 (UTC+8) | - |
| dc.date.available | 24-Feb-2025 15:55:36 (UTC+8) | - |
| dc.date.issued (上傳時間) | 24-Feb-2025 15:55:36 (UTC+8) | - |
| dc.identifier.uri (URI) | https://ah.lib.nccu.edu.tw/item?item_id=175330 | - |
| dc.description.abstract (摘要) | Ultrasound has an invaluable role in assessing musculoskeletal conditions, particularly the knee joint. However, it requires a long learning time for junior sonographers to mature their scanning skills, which, thus, limits its routine use in medical care. This study developed a deep image guiding (DIG) system as a quantitative model to reduce the learning curve and make the scanning procedure more objective and consistent. A total of 2772 images, including seven standard plane and one nonstandard plane datasets: S_infra_p (337), S_lat_med (381), S_LFTJ (302), S_MFTJ (345), S_post (154), S_supra_p (398), S_trans (210), and Non_S (645), comprised the database that was used for model training. DIG was a two-stage classification. The first classification was based on the vision transformer (ViT) model to interpret long-range anatomical structures for the classification of the eight plane categories. The second classification predicted the plane that the operator was likely to acquire using probability features of the current nonstandard plane and retrieval features from deep retrieval, which provided comparative differences between a nonstandard plane and the other seven standard planes in representative ViT feature space. DIG achieved an accuracy of 91.12% in distinguishing standard planes and nonstandard planes. ViT features were significantly more accurate compared to DenseNet201. In the second classification, combining probability and retrieval features achieved 87% accuracy, which was significantly better than using probability features alone. With tenfold cross-validation, the support vector machine (SVM) performed the best in four machine learning classifiers. Using DIG helps to guide the junior sonographers when scanning the desired standard plane, which may shorten the learning time required for the junior sonographers to become experienced and also make the scanning procedure more consistent. | |
| dc.format.extent | 104 bytes | - |
| dc.format.mimetype | text/html | - |
| dc.relation (關聯) | IEEE Transactions on Instrumentation & Measurement, Vol.73, pp.1-9 | |
| dc.subject (關鍵詞) | Deep learning; image retrieval; knee ultrasound; standard planes; vision transformer (ViT) | |
| dc.title (題名) | Deep Image Guiding: Guide Knee Ultrasound Scanning using Hierarchical Classification and Retrieval | |
| dc.type (資料類型) | article | |
| dc.identifier.doi (DOI) | 10.1109/TIM.2024.3476515 | |
| dc.doi.uri (DOI) | https://doi.org/10.1109/TIM.2024.3476515 | |
