Survival Analysis for Multimode Ablation Using Self-Adapted Deep Learning Network Based on Multisource Features.

Area of Science:

• Oncology
• Medical Imaging
• Bioinformatics

Background:

• Thermal ablation, including freezing and radio-frequency heating, offers therapeutic benefits for liver cancer but faces challenges with tumor recurrence.
• Accurate monitoring of tumor progression post-ablation is crucial for effective patient management and treatment efficacy assessment.
• Existing survival prediction models may not fully leverage multimodal data, including advanced imaging and immune markers.

Purpose of the Study:

• To develop and validate a novel survival analysis framework for predicting survival and assessing efficacy after thermal ablation for liver cancer.
• To integrate preoperative and postoperative MRI radiomics, deep learning features, and immune response data for enhanced prediction accuracy.
• To improve the accuracy of individual survival probability prediction over time for clinical decision support.

Main Methods:

• Extraction of MRI radiomics and vision transformer-based deep learning features from patient scans.
• Collection of immune features from peripheral blood using flow cytometry and routine blood tests.
• Development of a survival analysis framework using random survival forest for feature selection and an improved deep Cox mixture (DCM) model with a self-adapted fully connected layer for multimodal data integration.

Main Results:

• The proposed framework achieved a high C-statistic (C$^{\mathit{td}}$-index) of 0.885 ± 0.040 and an integrated Brier score of 0.041 ± 0.014, outperforming existing methods.
• Immune features demonstrated the highest importance, significantly contributing to prediction accuracy.
• The model accurately predicted individual patient survival probabilities over time.

Conclusions:

• The novel survival analysis framework effectively integrates multimodal data for accurate survival prediction and efficacy assessment in liver cancer patients undergoing thermal ablation.
• The significant contribution of immune features highlights their importance in predicting outcomes after ablation.
• This approach provides clinicians with trustworthy prognostic suggestions, aiding in personalized treatment strategies.