Can machine learning models predict my treatment outcome for uterine fibroids?

Machine learning (ML) is a type of artificial intelligence that can analyze patterns in data to make predictions. For uterine fibroids, researchers are testing whether ML models can predict how well a treatment will work, such as how much of a fibroid will shrink after a procedure. The short answer is that ML models show promise, especially for high-intensity focused ultrasound (HIFU) ablation, but they are not yet accurate enough to replace your doctor's judgment. Most studies are still in early stages and need more testing before they can be used in everyday care.

What the research says

A 2025 systematic review and meta-analysis looked at 14 studies on ML models for predicting outcomes of minimally invasive fibroid treatments like HIFU and uterine artery embolization (UAE) ³. The review found that logistic regression was the most common approach, and some models using gradient boosting showed high accuracy in individual studies. However, external validation (testing the model on a separate group of patients) was rare, which means we don't know how well these models would work in different hospitals or populations ³.

One study of 1,000 women who had HIFU found that an XGBoost model (a type of ML) was best at predicting whether at least 80% of the fibroid would be ablated (destroyed). The model's AUC (a measure of accuracy) was 0.692, which is moderate but not high enough for confident predictions ⁴. Key factors that helped the model predict success included MRI features like T2-weighted image signal, the distance from the fibroid to the skin, and the patient's platelet count ⁴.

Another study of 73 women used ML to predict whether HIFU would achieve at least 90% ablation. The gradient boosting model performed best, with accuracy ranging from 0.34 to 0.97 depending on the features used ⁵. A larger study of 573 fibroids in 410 women used MRI-based ML models to predict both short-term shrinkage and long-term regrowth. The best model for predicting shrinkage had an R² of 0.851 (meaning it explained 85% of the variation), and the model for predicting regrowth had an AUC of 0.904, which is quite good ⁶.

Overall, the evidence suggests ML models can be helpful, but they are not yet ready for routine use. The models need to be tested on more diverse groups of patients and in real-world settings before doctors can rely on them to guide treatment decisions.

What to ask your doctor

• Are there any machine learning tools available at your practice to help predict my fibroid treatment outcome?
• What factors do you consider most important when deciding which treatment (like HIFU, UAE, or surgery) is best for my fibroids?
• How do you measure success after fibroid treatment, and what are the chances my fibroid will shrink or grow back?
• Should I have any specific MRI or blood tests before treatment to help predict my results?
• Are there any clinical trials or research studies I could join that use machine learning to personalize fibroid care?