Triantafyllou Matthaios, Vassalou Evangelia E, Klontzas Michail E, Tosounidis Theodoros H, Marias Kostas, Karantanas Apostolos H
Artificial Intelligence and Translational Imaging (ATI) Lab, Department of Radiology, School of Medicine, University of Crete, Voutes Campus, Heraklion, Greece.
Department of Medical Imaging, University Hospital of Heraklion, Voutes, 71110, Heraklion, Crete, Greece.
Jpn J Radiol. 2025 May;43(5):829-840. doi: 10.1007/s11604-024-01725-x. Epub 2025 Jan 3.
Calcific tendinopathy, predominantly affecting rotator cuff tendons, leads to significant pain and tendon degeneration. Although US-guided percutaneous irrigation (US-PICT) is an effective treatment for this condition, prediction of patient' s response and long-term outcomes remains a challenge. This study introduces a novel radiomics-based model to forecast patient outcomes, addressing a gap in the current predictive methodologies.
The study involved 84 patients who underwent US-PICT, with data collected on clinical and demographic factors, alongside radiomic features extracted from ultrasound images. Key radiomic features predictive of the outcome were discerned through Least Absolute Shrinkage and Selection Operator (LASSO) method. Machine Learning models, including Random Forest, XGBoost, and Support Vector Machines, were employed to analyze the radiomics, the clinical and the combined dataset, focusing on calcium removal extent. An external testing was conducted using an independent cohort from a different institution to assess the model's generalizability. Metrics were calculated for the best-performing models, namely area under the curve (AUC) score, sensitivity, specificity, precision or positive predictive value, and negative predictive value.
The selected features were merged with clinical data, notably the calcification's maximum diameter. This enriched dataset was fed into classification models. The superior model achieved an AUC of 0.88 (95% CI 0.73-0.99), with a positive predictive value of 0.92 and a sensitivity of 0.90. In external testing, the combined model achieved an AUC of 0.78. SHAP analysis was employed to highlight the impact of the selected features on the optimal model's effectiveness.
The developed radiomics model offers a promising tool for predicting outcomes of US-PICT, potentially guiding clinical decision-making.
钙化性肌腱病主要影响肩袖肌腱，会导致严重疼痛和肌腱退变。尽管超声引导下经皮冲洗（US-PICT）是治疗这种疾病的有效方法，但预测患者的反应和长期预后仍然是一项挑战。本研究引入了一种基于放射组学的新型模型来预测患者预后，以填补当前预测方法的空白。
该研究纳入了84例行US-PICT的患者，收集了临床和人口统计学因素数据，以及从超声图像中提取的放射组学特征。通过最小绝对收缩和选择算子（LASSO）方法识别出预测预后的关键放射组学特征。采用包括随机森林、XGBoost和支持向量机在内的机器学习模型分析放射组学、临床和综合数据集，重点关注钙清除程度。使用来自不同机构的独立队列进行外部测试，以评估模型的可推广性。为表现最佳的模型计算指标，即曲线下面积（AUC）评分、敏感性、特异性、精确率或阳性预测值以及阴性预测值。
所选特征与临床数据（尤其是钙化的最大直径）合并。这个丰富的数据集被输入到分类模型中。最佳模型的AUC为0.88（95%CI 0.73 - 0.99），阳性预测值为0.92，敏感性为0.90。在外部测试中，联合模型的AUC为0.78。采用SHAP分析来突出所选特征对最佳模型有效性的影响。
所开发的放射组学模型为预测US-PICT的预后提供了一个有前景的工具，可能会指导临床决策。
