Ultrasound Radiomics Correlating With Clinical Markers for Enhanced Detection of Placenta Accreta Spectrum

Area of Science:

• Radiology and Medical Imaging
• Obstetrics and Gynecology
• Artificial Intelligence in Medicine

Background:

• Placenta accreta spectrum (PAS) poses significant risks during pregnancy, necessitating accurate diagnostic methods.
• Existing sonographic markers for PAS can be subjective and may not always provide definitive diagnoses.
• Texture analysis in ultrasound imaging offers potential for objective quantitative assessment.

Purpose of the Study:

• To develop a quantitative ultrasound index for diagnosing placenta accreta spectrum (PAS) by integrating B-mode and color Doppler texture features.
• To evaluate the correlation between the developed index and established sonographic markers for PAS.
• To assess the diagnostic performance of B-mode, color Doppler, and multimodal ultrasound-based radiomics models for PAS detection.

Main Methods:

• Retrospective analysis of 2106 texture features from mid-trimester B-mode and Doppler ultrasound images of patients with placenta previa and prior uterine surgery risk factors.
• Development of three distinct radiomics models: B-mode only, color Doppler only, and a multimodal approach integrating both.
• Feature selection using a pipeline combining linear discriminant analysis, extra trees regression, and recursive feature elimination, followed by five-fold cross-validation and testing on a hold-out set.

Main Results:

• The multimodal model achieved 90% (±6.7%) accuracy in detecting PAS, outperforming the B-mode (88.7% ±5.3%) and color Doppler (85.1% ±9.1%) models.
• All evaluated clinical disease markers were discriminative of the quantitative index generated from B-mode and multimodal models.
• The multimodal approach demonstrated superior diagnostic capability compared to individual modality models.

Conclusions:

• Combined B-mode and Doppler ultrasound-based radiomics provide an accurate and computationally inexpensive method for detecting PAS in high-risk patients.
• The multimodal index captures critical diagnostic information potentially missed by single-modality assessments.
• These models show potential as valuable aids for clinical decision-making during routine mid-trimester scans, exceeding projected clinical standards.