Multimodalidad y análisis temporal de la respuesta al tratamiento del cáncer de cuello uterino

Cervical cancer remains a significant global health challenge, requiring improved diagnostic and prognostic tools to enhance treatment planning and outcomes. Noninvasive medical imaging offers a promising route for precision diagnostics. This study aimed to develop and evaluate a predictive model for cervical cancer treatment response during radiation therapy using features extracted from multimodal medical imaging.

This study evaluated the use of multimodal medical imaging, including apparent diffusion coefficient (ADC), dynamic contrast-enhanced (DCE), and positron emission tomography (PET), across different treatment stages (pre-, mid-, and post-stage) in 22 patients with cervical cancer. From these images, we extracted and assessed the predictive performance of various feature types, including zero-order, first-order, second-order, and higher-order features.

Texture features, particularly those derived from the Gray Level Co-occurrence Matrix (GLCM) in two-dimensional (2D) plane, were more effective compared to other image features, achieving an area under the curve (AUC) of 0.73±0.12. Combining GLCM with shape features further increased the AUC to 0.75. Among the GLCM features, "contrast" was identified as the most predictive for treatment response (AUC of 0.74 for the top five contrast features). Among single-modality analyses, ADC demonstrated the best prediction compared to PET/computed tomography (CT) (15% AUC increase) and DCE (12% AUC increase). The combination of imaging modalities and texture analysis further enhanced patient stratification, yielding an average 8% AUC increase compared to single-modality models. Using only post-stage GLCM2D features resulted in an AUC only 4% lower than using all time points, suggesting that reduced imaging time points and modalities may still retain strong predictive power.

Integrating texture features from multimodal imaging can improve cervical cancer prognostication and guide personalized treatment strategies. These findings support the potential of imaging biomarkers in optimizing therapy and reducing the diagnostic burden, contributing to more efficient and tailored cancer care.