Scholarlink

Objective <p>To investigate the utility of a machine learning-based multi-parameter magnetic resonance imaging (MRI) radiomics model in the identification of partial anterior cruciate ligament (ACL) tears.</p> Methods <p>Each scan comprised T1-weighted imaging (T1WI) and proton density-weighted imaging (PDWI). The entire intercondylar fossa was delineated as the region of interest, from which 1,045 radiomics features were independently extracted. A random division of all cases into training and testing cohorts was performed in a 7:3 ratio. Feature selection was executed via Spearman correlation analysis, the least absolute shrinkage and selection operator, and recursive feature elimination. Utilizing the optimal features, three models—T1, PD, and T1&amp;PD—were established employing a support vector machine algorithm. Model performance was evaluated by metrics encompassing accuracy, precision, recall, F1-score, and receiver operating characteristic (ROC) curve analysis.</p> Results <p>A subset of 54 radiological features was retained from 2090, consisting of 30 from T1WI and 24 from PDWI. Three models (T1, PD, and T1&amp;PD) were developed based on these selected features. Within the training cohort, the T1&amp;PD model yielded precision, recall, and F1-scores of 0.822, 0.909, and 0.863, respectively. Its performance substantially exceeded that of the single-sequence T1 and PD models (<i>P</i> &lt; 0.001). ROC curve analysis in the testing cohort indicated that the combined T1&amp;PD model demonstrated the highest classification efficacy, with area under the curve values of 0.887, 0.889, and 0.847 for no tear, partial tear, and complete tear, respectively.</p> Conclusion <p>The machine learning-driven multi-parameter MRI radiomics model demonstrated favorable effectiveness in distinguishing partial ACL tears.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Multi-parameter MRI radiomics model based on machine learning for identifying partial tears of the anterior cruciate ligament

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年4月6日

查看全文
  • Qi Cheng,
  • Pengfei Zhu,
  • Weiming Cai,
  • Yuan Liu,
  • Han Wang,
  • Lijie Shi

摘要

Objective

To investigate the utility of a machine learning-based multi-parameter magnetic resonance imaging (MRI) radiomics model in the identification of partial anterior cruciate ligament (ACL) tears.

Methods

Each scan comprised T1-weighted imaging (T1WI) and proton density-weighted imaging (PDWI). The entire intercondylar fossa was delineated as the region of interest, from which 1,045 radiomics features were independently extracted. A random division of all cases into training and testing cohorts was performed in a 7:3 ratio. Feature selection was executed via Spearman correlation analysis, the least absolute shrinkage and selection operator, and recursive feature elimination. Utilizing the optimal features, three models—T1, PD, and T1&PD—were established employing a support vector machine algorithm. Model performance was evaluated by metrics encompassing accuracy, precision, recall, F1-score, and receiver operating characteristic (ROC) curve analysis.

Results

A subset of 54 radiological features was retained from 2090, consisting of 30 from T1WI and 24 from PDWI. Three models (T1, PD, and T1&PD) were developed based on these selected features. Within the training cohort, the T1&PD model yielded precision, recall, and F1-scores of 0.822, 0.909, and 0.863, respectively. Its performance substantially exceeded that of the single-sequence T1 and PD models (P < 0.001). ROC curve analysis in the testing cohort indicated that the combined T1&PD model demonstrated the highest classification efficacy, with area under the curve values of 0.887, 0.889, and 0.847 for no tear, partial tear, and complete tear, respectively.

Conclusion

The machine learning-driven multi-parameter MRI radiomics model demonstrated favorable effectiveness in distinguishing partial ACL tears.