Cervical cancer is a major public health concern that poses a significant threat to women's health. Standard concurrent chemoradiotherapy (CCRT) achieves disease-free survival (DFS) in approximately 70% of patients with locally advanced disease, however, nearly 30% still experience recurrence or metastasis.

Intensified treatment strategies may improve survival rates, but they often come with higher toxicity and costs. The key challenge is to accurately identify patients who truly need intensive treatment by the clinicians, as standard therapy suffices for low-risk cases, whereas high-risk patients benefit from aggressive intervention.

In a study published in npj Digital Medicine, a team led by Assoc. Prof. LIANG Xiaokun from the Shenzhen Institute of Advanced Technology of the Chinese Academy of Sciences, along with Prof. HU Ke and Prof. HOU Xiaorong from the Peking Union Medical College Hospital, developed a deep learning-based multimodal prognostic prediction model, CerviPro, which provides a comprehensive approach for risk stratification, enabling personalized treatment strategies.

CerviPro integrates features of pre- and post-radiotherapy CT foundation model, radiomics features, and clinical information. It first employs deep learning-based automatic segmentation techniques to precisely extract tumor regions, and then utilizes a pre-trained CT foundation model to extract high-dimensional deep features. Subsequently, intelligent fusion of multi-source heterogeneous data is achieved through principal component analysis for dimensionality reduction and feature selection techniques.

To ensure the clinical applicability of CerviPro, researchers collected multimodal clinical data from 1,018 cervical cancer patients across multiple hospitals in China. Using a multi-center validation design, they demonstrated that the model not only achieved high performance in individual hospital settings, but also maintained robustness and adaptability across diverse real-world clinical environments.

CerviPro achieved robust predictive performance across all testing cohorts (training, internal validation, and external validation), and demonstrated superior predictive performance compared to conventional Cox proportional hazards models and DeepSurv. It successfully stratified patients into high-risk DFS group which potentially requires intensified treatment and low-risk DFS group which may consider de-escalated treatment options, along with other critical prognostic insights.

This study offers clinicians a reliable, intelligent decision-support tool for accurate identification of high-risk patients, guiding the development of personalized treatment strategies for locally advanced cervical cancer.