MRI and Clinical Features for Deep Infiltrating Endometriosis Diagnosis

330 patient records with surgical-pathological confirmation were collected from January 2020 to June 2025. Linlin Zhang developed a weighted ensemble machine learning model using multimodal MRI and clinical features to diagnose deep infiltrating endometriosis. The model achieved an AUC of 0.938 on an independent test set and outperformed radiologists' average AUC of 0.818.

Use Cases

• Training diagnostic AI models based on multimodal MRI features mentioned in the description
• Comparing machine learning model performance against human expert readings based on the study's validation approach
• Investigating feature importance for endometriosis diagnosis based on the SHAP analysis identifying CRP, CA125, T2-weighted signal intensity, and symptom duration

Strengths

• 330 patient records with surgical-pathological confirmation provide a validated clinical ground truth
• Model performance is quantified with an AUC of 0.938, sensitivity of 0.920, and accuracy of 0.859 on an independent test set
• The study includes a comparison against radiologists' performance (average AUC 0.818) and measures AI-assisted improvement (accuracy increased to 0.920)

Limitations

• Column-level documentation is absent; field semantics must be inferred after download
• Row count is unknown, which may limit suitability assessment
• The 11.5 KB file size suggests the dataset likely contains summary or model data, not the full raw imaging or clinical records

Provenance

Source

figshare

Collection Method

Retrospective study enrolling patients who underwent MRI examinations and surgical-pathological confirmation.

Time Range

January 2020 to June 2025

Freshness

Last updated 2026-05-29 04:44:31