Department of Musculoskeletal Tumor, The Third Hospital of Hebei Medical University, Shijiazhuang, China Shijiazhuang, China.
Eur Rev Med Pharmacol Sci. 2023 Jun;27(11):5039-5052. doi: 10.26355/eurrev_202306_32621.
Giant cell tumor of bone (GCTB) is a common primary bone tumor with latent malignant tendency. GCTB is prone to occur around the knee joint, and surgery is the major treatment method. There are relatively few reports on denosumab in the treatment of recurrent GCTB around the knee joint and postoperative function evaluation of patients. This research aimed to explore the appropriate surgical options for the treatment of recurrent GCTB around the knee joint.
19 patients with recurrent GCTB around the knee joint, who were admitted to Hospital for 3 months following denosumab treatment from January 2016 to December 2019, were included as the research subjects. The prognosis was compared between patients treated with curettage combined with polymethylmethacrylate (PMMA) and those with extensive-resection replacement of tumor prosthesis (RTP). A deep learning model of Inception-v3 combined with a Faster region-based convolutional neural network (Faster-RCNN) was constructed to classify and identify X-ray images of patients. The Musculoskeletal Tumor Society (MSTS) score, short form-36 (SF-36) score, recurrence, and the rate of complications were also analyzed during the follow-up period.
The results showed that the Inception-v3 model trained on the low-rank sparse loss function was obviously the best for X-ray image classification, and the classification and identification effect of the Faster-RCNN model was significantly better than that of the convolutional neural network (CNN), U-Net, and Fast region-based convolutional neural network (Fast-RCNN) models. During the follow-up period, the MSTS score in the PMMA group was significantly higher than that in the RTP group (p<0.05), while there was no significant difference in the SF-36 score, recurrence, and the rate of complications (p>0.05).
The deep learning model could improve the classification and identification of the lesion location in the X-ray images of GCTB patients. Denosumab was an effective adjuvant for recurrent GCTB, and widely extensive-resection RTP could reduce the risk of local recurrence after denosumab treatment for recurrent GCTB.
骨巨细胞瘤（GCTB）是一种具有潜在恶性倾向的常见原发性骨肿瘤。GCTB 易发生于膝关节周围，手术是主要治疗方法。关于 denosumab 治疗膝关节周围复发性 GCTB 及患者术后功能评估的报道相对较少。本研究旨在探讨治疗膝关节周围复发性 GCTB 的合适手术方案。
2016 年 1 月至 2019 年 12 月，共收治 19 例接受 denosumab 治疗后 3 个月内复发的膝关节周围 GCTB 患者。比较刮除联合聚甲基丙烯酸甲酯（PMMA）与广泛切除肿瘤假体置换（RTP）治疗的疗效。构建了 Inception-v3 与 Faster 区域卷积神经网络（Faster-RCNN）相结合的深度学习模型，对患者的 X 射线图像进行分类和识别。分析随访期间的 MSTS 评分、SF-36 评分、复发率和并发症发生率。
结果表明，基于低秩稀疏损失函数训练的 Inception-v3 模型对 X 射线图像分类效果最佳，Faster-RCNN 模型的分类和识别效果明显优于卷积神经网络（CNN）、U-Net 和 Fast 区域卷积神经网络（Fast-RCNN）模型。随访期间，PMMA 组的 MSTS 评分明显高于 RTP 组（p<0.05），SF-36 评分、复发率和并发症发生率无显著差异（p>0.05）。
深度学习模型可提高 GCTB 患者 X 射线图像中病变部位的分类和识别能力。Denosumab 是复发性 GCTB 的有效辅助治疗药物，广泛广泛切除 RTP 可降低 denosumab 治疗复发性 GCTB 后局部复发的风险。
Zotero 插件