Three-dimensional MRI follicle segmentation and counting using SegmentWithSAM in the diagnosis of polycystic ovary syndrome

Abstract

Objective

This study aimed to investigate the diagnostic performance of Follicle numbers measured on ultrasound (US), conventional magnetic resonance imaging (2D MRI), and three-dimensional (3D) MRI in patients with polycystic ovary syndrome (PCOS) and to compare the diagnostic efficacy of these imaging modalities.

Method

In this prospective study, 58 PCOS patients and 60 healthy women underwent US, conventional 2D MRI, and 3D MRI. Clinical laboratory tests and ovarian volume were compared between PCOS and control groups. Follicle numbers measured on US (FN-US), 2D MRI (FN-2D), and 3D MRI (FN-3D) using SegmentWithSAM were compared between PCOS and control groups using receiver operating characteristic (ROC) curve analysis and the DeLong test.

Results

Ovarian volume and follicle numbers were significantly higher in the PCOS group than in the control group. The diagnostic performance was found with FN-3D achieving the highest AUC of 0.94 (95% CI: 0.90–0.98), superior to that of US (0.80 [95% CI: 0.72–0.88]) and 2D MRI (0.90 [95% CI: 0.84–0.96]), respectively. Significant differences in the diagnostic efficacy of follicle counts were observed between US, conventional MRI, and 3D MRI, with 3D MRI showing superior results.

Conclusion

3D MRI was superior to US and 2D MRI in diagnosing PCOS, with the ability to display small follicles.

Graphical abstract