A novel accumulative modelling method using temporal images: Concept proposal and clinical validation on multiple...

Area of Science:

• Radiotherapy and Medical Imaging
• Radiomics and Computational Pathology
• Predictive Modeling in Oncology

Background:

• Image-guided radiotherapy (IGRT) relies on serial imaging for treatment adaptation.
• Temporal changes in imaging features can indicate treatment response or toxicity.
• Predictive models are crucial for personalizing cancer treatment and minimizing adverse events.

Purpose of the Study:

• To introduce and validate a novel accumulative radiomics prediction method using temporal data from IGRT.
• To assess the performance of this method in predicting radiation pneumonitis (RP) and pathologic complete response (pCR).
• To compare the proposed method against conventional approaches and evaluate its generalizability.

Main Methods:

• Calculated temporal changes of 560 radiomics features from Cone-Beam CT (CBCT) scans (CBCT0, CBCTi).
• Developed an accumulative delta-radiomics feature (Delta-RFaccu) by stacking temporal features.
• Compared prediction performance using univariate and multivariate logistic regression, incorporating clinical and dosimetric predictors.

Main Results:

• The Delta-RFaccu signature achieved an AUC of 0.82 ± 0.09 for RP prediction, outperforming conventional Delta-RF signatures.
• Performance improvements were also observed in pCR prediction, demonstrating reproducibility.
• Integrating Delta-RFaccu with clinical/dosimetric factors further increased AUC to 0.85 ± 0.10 (RP) and 0.87 ± 0.06 (pCR).

Conclusions:

• A novel accumulative modeling method using temporal imaging data demonstrated superior predictive ability across multiple clinical tasks.
• The proposed method, integrated into clinical models, can aid in patient-specific decision-making for radiotherapy.
• This approach has the potential to reduce radiation-induced injuries and optimize surgical interventions.