Scholarlink

Objective <p>Testicular volume serves as a crucial indicator for evaluating therapeutic outcomes in patients with congenital hypogonadotropic hypogonadism (CHH), which is a rare disease. We aim to develop a multimodal machine learning model that integrates clinical data, testicular ultrasonography, and pituitary magnetic resonance imaging (MRI) radiomic features to predict whether testicular volume exceeds 4 mL following treatment in CHH patients.</p> Methods <p>This study retrospectively included 66 male patients with CHH from 2015 to 2024, with a follow-up period of 12 to 36 months. All patients received gonadotropin or GnRH pulse pump therapy. The patients’ visit information was systematically collected, and the peak testicular volume during follow-up was used as the outcome. Multiple machine learning prediction models were constructed using clinical information, initial-visit testicular ultrasound images, and pituitary MRI images, and were evaluated using ten-fold cross-validation. Model performance was compared using the Delong test. Model interpretability was analyzed using SHapley Additive exPlanations (SHAP).</p> Results <p>A total of 28 patients achieved a testicular volume greater than 4 mL after treatment, indicating favorable reproductive outcomes. The multimodal model that integrated all information demonstrated the highest predictive performance, with an area under the receiver operating characteristic curve (AUC) of 0.932 (95% CI: 0.891–0.997), outperforming the clinical model (AUC = 0.863, 95% CI: 0.747–1.000). The best AUCs for models based solely on ultrasound and MRI radiomic features were 0.798 (95% CI: 0.697–1.000) and 0.680 (95% CI: 0.526–0.868), respectively. The combined imaging model (ultrasound + MRI) yielded an AUC of 0.759 (95% CI: 0.510–0.969). Among the top 10 SHAP-ranked features in the multimodal model, two originated from ultrasound, one from MRI, and seven from clinical variables.</p> Conclusion <p>The multimodal model, which integrated clinical factors with radiomic features from both ultrasound and magnetic resonance imaging, demonstrated favorable performance in predicting reproductive outcomes in patients with CHH after treatment.</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

A multimodal machine learning model for predicting reproductive outcomes in the rare congenital hypogonadotropic hypogonadism after treatment

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年8月4日

查看全文
  • Xiaomeng Li,
  • Chuyin Ruan,
  • Zebin Wu,
  • Sina Wang,
  • Tianli Liang,
  • Derun Pan,
  • Ying Cao,
  • Weiguo Chen

摘要

Objective

Testicular volume serves as a crucial indicator for evaluating therapeutic outcomes in patients with congenital hypogonadotropic hypogonadism (CHH), which is a rare disease. We aim to develop a multimodal machine learning model that integrates clinical data, testicular ultrasonography, and pituitary magnetic resonance imaging (MRI) radiomic features to predict whether testicular volume exceeds 4 mL following treatment in CHH patients.

Methods

This study retrospectively included 66 male patients with CHH from 2015 to 2024, with a follow-up period of 12 to 36 months. All patients received gonadotropin or GnRH pulse pump therapy. The patients’ visit information was systematically collected, and the peak testicular volume during follow-up was used as the outcome. Multiple machine learning prediction models were constructed using clinical information, initial-visit testicular ultrasound images, and pituitary MRI images, and were evaluated using ten-fold cross-validation. Model performance was compared using the Delong test. Model interpretability was analyzed using SHapley Additive exPlanations (SHAP).

Results

A total of 28 patients achieved a testicular volume greater than 4 mL after treatment, indicating favorable reproductive outcomes. The multimodal model that integrated all information demonstrated the highest predictive performance, with an area under the receiver operating characteristic curve (AUC) of 0.932 (95% CI: 0.891–0.997), outperforming the clinical model (AUC = 0.863, 95% CI: 0.747–1.000). The best AUCs for models based solely on ultrasound and MRI radiomic features were 0.798 (95% CI: 0.697–1.000) and 0.680 (95% CI: 0.526–0.868), respectively. The combined imaging model (ultrasound + MRI) yielded an AUC of 0.759 (95% CI: 0.510–0.969). Among the top 10 SHAP-ranked features in the multimodal model, two originated from ultrasound, one from MRI, and seven from clinical variables.

Conclusion

The multimodal model, which integrated clinical factors with radiomic features from both ultrasound and magnetic resonance imaging, demonstrated favorable performance in predicting reproductive outcomes in patients with CHH after treatment.