1
|
Xiong Y, Guo W, Liang Z, Wu L, Ye G, Liang YY, Wen C, Yang F, Chen S, Zeng XW, Xu F. Deep learning-based diagnosis of osteoblastic bone metastases and bone islands in computed tomograph images: a multicenter diagnostic study. Eur Radiol 2023; 33:6359-6368. [PMID: 37060446 PMCID: PMC10415522 DOI: 10.1007/s00330-023-09573-5] [Citation(s) in RCA: 0] [Impact Index Per Article: 0] [Reference Citation Analysis] [Abstract] [Key Words] [MESH Headings] [Grants] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Received: 02/04/2023] [Revised: 03/08/2023] [Accepted: 03/20/2023] [Indexed: 04/16/2023]
Abstract
OBJECTIVE To develop and validate a deep learning (DL) model based on CT for differentiating bone islands and osteoblastic bone metastases. MATERIALS AND METHODS The patients with sclerosing bone lesions (SBLs) were retrospectively included in three hospitals. The images from site 1 were randomly assigned to the training (70%) and intrinsic verification (10%) datasets for developing the two-dimensional (2D) DL model (single-slice input) and "2.5-dimensional" (2.5D) DL model (three-slice input) and to the internal validation dataset (20%) for evaluating the performance of both models. The diagnostic performance was evaluated using the internal validation set from site 1 and additional external validation datasets from site 2 and site 3. And statistically analyze the performance of 2D and 2.5D DL models. RESULTS In total, 1918 SBLs in 728 patients in site 1, 122 SBLs in 71 patients in site 2, and 71 SBLs in 47 patients in site 3 were used to develop and test the 2D and 2.5D DL models. The best performance was obtained using the 2.5D DL model, which achieved an AUC of 0.996 (95% confidence interval [CI], 0.995-0.996), 0.958 (95% CI, 0.958-0.960), and 0.952 (95% CI, 0.951-0.953) and accuracies of 0.950, 0.902, and 0.863 for the internal validation set, the external validation set from site 2 and site 3, respectively. CONCLUSION A DL model based on a three-slice CT image input (2.5D DL model) can improve the prediction of osteoblastic bone metastases, which can facilitate clinical decision-making. KEY POINTS • This study investigated the value of deep learning models in identifying bone islands and osteoblastic bone metastases. • Three-slice CT image input (2.5D DL model) outweighed the 2D model in the classification of sclerosing bone lesions. • The 2.5D deep learning model showed excellent performance using the internal (AUC, 0.996) and two external (AUC, 0.958; AUC, 0.952) validation sets.
Collapse
Affiliation(s)
- Yuchao Xiong
- Department of Radiology, Guangzhou Red Cross Hospital (Guangzhou Red Cross Hospital, Medical College of Jinan University), 396 Tongfu Road, Guangzhou, 510220, Guangdong Province, China
| | - Wei Guo
- Department of Radiology, Wuhan Third Hospital, Tongren Hospital of Wuhan University, 241 Liuyang Road, Wuhan, 430063, Hubei Province, China
| | - Zhiping Liang
- Department of Radiology, Guangzhou Red Cross Hospital (Guangzhou Red Cross Hospital, Medical College of Jinan University), 396 Tongfu Road, Guangzhou, 510220, Guangdong Province, China
| | - Li Wu
- Department of Radiology, Guangzhou Red Cross Hospital (Guangzhou Red Cross Hospital, Medical College of Jinan University), 396 Tongfu Road, Guangzhou, 510220, Guangdong Province, China
| | - Guoxi Ye
- Department of Radiology, Guangzhou Red Cross Hospital (Guangzhou Red Cross Hospital, Medical College of Jinan University), 396 Tongfu Road, Guangzhou, 510220, Guangdong Province, China
| | - Ying-Ying Liang
- Department of Radiology, Guangzhou First People's Hospital, School of Medicine, South China University of Technology, 1Panfu Road, Guangzhou, 510180, Guangdong Province, China
| | - Chao Wen
- Department of Radiology, Guangzhou Red Cross Hospital (Guangzhou Red Cross Hospital, Medical College of Jinan University), 396 Tongfu Road, Guangzhou, 510220, Guangdong Province, China
| | - Feng Yang
- Department of Radiology, Guangzhou Red Cross Hospital (Guangzhou Red Cross Hospital, Medical College of Jinan University), 396 Tongfu Road, Guangzhou, 510220, Guangdong Province, China
| | - Song Chen
- Department of Radiology, Guangzhou Red Cross Hospital (Guangzhou Red Cross Hospital, Medical College of Jinan University), 396 Tongfu Road, Guangzhou, 510220, Guangdong Province, China
| | - Xu-Wen Zeng
- Department of Radiology, Guangzhou Red Cross Hospital (Guangzhou Red Cross Hospital, Medical College of Jinan University), 396 Tongfu Road, Guangzhou, 510220, Guangdong Province, China.
| | - Fan Xu
- Department of Radiology, Guangzhou Red Cross Hospital (Guangzhou Red Cross Hospital, Medical College of Jinan University), 396 Tongfu Road, Guangzhou, 510220, Guangdong Province, China.
| |
Collapse
|
2
|
CT Attenuation Values Do Not Reliably Distinguish Benign Sclerotic Lesions From Osteoblastic Metastases in Patients Undergoing Bone Biopsy. AJR Am J Roentgenol 2021; 216:1022-1030. [PMID: 32755207 DOI: 10.2214/ajr.20.24029] [Citation(s) in RCA: 5] [Impact Index Per Article: 1.7] [Reference Citation Analysis] [Abstract] [Key Words] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Indexed: 12/21/2022]
Abstract
BACKGROUND. CT attenuation thresholds that accurately distinguish enostoses from untreated osteoblastic metastases have been published. In the Mayo Clinic practices, these thresholds have been applied more broadly to distinguish benign sclerotic bone lesions other than enostoses from osteoblastic metastases. OBJECTIVE. The purpose of this article is to determine if CT attenuation thresholds allow the distinguishing of benign sclerotic bone lesions from osteoblastic metastases in patients undergoing bone biopsy. METHODS. A retrospective search was conducted to identify sclerotic lesions described on CT between October 7, 1998, and July 15, 2018, that underwent subsequent biopsy. Two musculoskeletal radiologists recorded lesions' maximum and mean attenuation. Using previously published attenuation thresholds, sensitivity and specificity for differentiating benign sclerotic lesions from osteoblastic metastases were calculated. ROC curve analysis was performed to determine if more appropriate attenuation thresholds exist. Intraclass correlation coefficients (ICCs) were computed. RESULTS. A total of 280 patients met inclusion criteria. Of those, 162 had malignant biopsy results and 118 had benign biopsy results. Of the 162 malignant lesions, 81 had received prior treatment. Maximum and mean attenuation were not significantly different between benign and malignant lesions for either reader (all p > .05). For reader 1, to distinguish benign from malignant lesions, a maximum attenuation threshold of more than 1060 HU resulted in sensitivity of 23.7%, specificity of 87.0%, and accuracy of 60.6%. A mean attenuation threshold of greater than 885 HU resulted in sensitivity of 19.5%, specificity of 90.7%, and accuracy 60.7%. ROC curve analysis showed AUCs for mean and maximum attenuation thresholds of 51.8% and 54.6%, respectively. Subgroup analyses of benign versus malignant and treated versus untreated lesions had similar results. Similar findings were obtained for reader 2. The two readers' ICC was 0.946 for maximum attenuation and 0.918 for mean attenuation. CONCLUSION. Published attenuation thresholds for distinguishing enostoses from osteoblastic metastases had slightly decreased specificity and markedly decreased sensitivity when applied to the differentiation of benign sclerotic lesions from osteoblastic metastases in our sample of biopsy-proven lesions. ROC analysis showed no high-performing attenuation threshold alternative. CLINICAL IMPACT. Published CT attenuation thresholds intended for distinguishing enostoses from osteoblastic metastases should not be used more broadly. More accurate alternative thresholds could not be derived.
Collapse
|