Multimodal machine learning for menopause status prediction using LLM-extracted ultrasound features
-
By
-
Weiwei Yin
-
Zhengyuan Shen
-
Chun Feng
-
Xia Zhang
-
Sihao Shen
-
Yiyue Jiang
-
Zhenbo Cheng
-
Lihui Wang
-
Ling Liu
-
July 2, 2026
-
Clinical Scorecard: Integrating Multimodal Machine Learning for Predicting Menopausal Status through Ultrasound Features Extracted by LLM
At a Glance
| Category | Detail |
|---|
| Condition | Menopausal Status Assessment |
| Key Mechanisms | Integration of ultrasound morphological features with anthropometric and hormone data for predictive modeling. |
| Target Population | Women undergoing menopausal assessment, particularly those with atypical manifestations or missing hormone data. |
| Care Setting | Clinical scenarios involving gynecological examinations and hormone assessments. |
Key Highlights
- Study included 997 women for training and validation of predictive models.
- Highest AUC of 0.984 achieved by integrating anthropometric and hormone features.
- Ultrasound morphological features can enhance menopausal status prediction.
- LLM effectively extracts structured features from unstructured ultrasound reports.
- Predictive model maintains accuracy even with missing feature types.
Guideline-Based Recommendations
Diagnosis
- Hormone assessment is the diagnostic standard for menopause assessment.
- Use of FSH and E2 as key endocrine markers is recommended.
Management
- Initiate hormone replacement therapy within the optimal treatment window.
Monitoring & Follow-up
- Regular assessment of hormonal levels and ultrasound features during perimenopause.
Risks
- Long-term estrogen deficiency can lead to osteoporosis and cardiovascular diseases.
Patient & Prescribing Data
Women experiencing perimenopause or menopause.
Integration of ultrasound data can improve diagnostic confidence and treatment decisions.
Clinical Best Practices
- Combine hormonal assessments with ultrasound morphological evaluations for comprehensive menopausal assessment.
- Utilize LLM for structured feature extraction from ultrasound reports.
Related Resources & Content
