Delta Radiomics and Tumor Size: A New Predictive Radiomics Model for Chemotherapy Response in Liver Metastases from Breast and Colorectal Cancer

Nicolò Gennaro, Moataz Soliman, Amir A. Borhani, Linda C. Kelahan, Hatice Savas, Ryan Avery, Kamal Subedi, Tugce Agirlar Trabzonlu, Chase Krumpelman, Vahid Yaghmai, Young Kwang Chae, Jochen H. Lorch, Devalingam Mahalingam, Mary F. Mulcahy, Al B. Benson, Ulaş Bağcı, Yury VelichkoTomography 2025 · DOI

TL;DR

This paper introduces "baseline-referenced Delta radiomics," a novel approach that integrates radiomic features with tumor size to predict chemotherapy response in liver metastases. This new model shows comparable or superior performance to existing methods, offering a promising tool for personalized cancer treatment.

What Problem Does This Paper Address?

Liver metastases from breast and colorectal cancer are a significant clinical challenge, often requiring systemic chemotherapy when surgical resection isn't an option. Accurately assessing a patient's response to chemotherapy early on is crucial for optimizing treatment strategies and improving patient outcomes. Traditional methods like RECIST 1.1 rely on tumor size changes, which can be a limited indicator of treatment efficacy. Radiomics, which extracts quantitative features from medical images, offers a more detailed view of tumor characteristics. However, existing radiomics approaches often require both pre- and post-treatment scans, and the relationship between radiomic features and tumor size, especially after treatment, is complex and not fully understood. This study aims to bridge this gap by developing a more robust radiomics model that accounts for the interplay between radiomic features and tumor size to predict chemotherapy response.

Methods

The researchers developed a novel "baseline-referenced Delta radiomics" model. This approach leverages post-treatment CT scans to assess therapy response. Instead of directly using pre-treatment radiomic features, it utilizes a "baseline reference function" derived from the relationship between radiomic features and tumor size observed in large pre-treatment datasets. This function essentially normalizes post-treatment radiomic features based on a patient's individual baseline characteristics, allowing for a more personalized assessment of change.

The study involved several key steps:

  1. CT Preprocessing: Standardized image processing techniques were applied.
  2. Tumor Segmentation: Liver metastases were manually segmented by radiologists.
  3. Radiomic Feature Extraction: A comprehensive set of radiomic features (intensity-based, shape, texture, etc.) were extracted using LIFEx software.
  4. Radiomic Feature Harmonization: A linear mixed-effects model (LME) was used to account for variability introduced by different CT scanners, ensuring consistency across datasets.
  5. Model Building and Statistical Analysis: Three models were compared:
  6. Pretreatment Radiomics Model: Used only baseline radiomic features.
  7. Delta Radiomics Model: Used the difference between pre- and post-treatment radiomic features.
  8. Baseline-Referenced Delta Radiomics Model: The novel approach, using post-treatment scans and the derived baseline reference functions.

Machine learning algorithms, including Random Forest, LogitBoost, svmRadial, and pcaNNet, were employed to build classification models for predicting response. Model performance was evaluated using confusion matrices, sensitivity, specificity, balanced accuracy, and ROC curves.

Datasets

The study utilized retrospective contrast-enhanced computed tomography (CT) scans from two patient cohorts:

  • Breast Cancer (BC): 83 patients with 155 liver metastases for initial dataset harmonization, and a subset of 57 patients with 106 lesions for response assessment.
  • Colorectal Cancer (CRC): 84 patients with 192 liver metastases for initial dataset harmonization, and a subset of 37 patients with 109 lesions for response assessment.

Additionally, a dataset of 228 patients with 280 hepatic cysts and 707 samples of normal liver parenchyma was used for radiomics data harmonization. The study did not introduce new benchmark datasets but rather analyzed existing patient data.

Key Insights and Ideas

  • Tumor Size and Radiomics are Intertwined: The study highlights that radiomic features are not independent of tumor size, especially before treatment. This relationship changes after treatment due to therapy efficacy.
  • Baseline Reference is Key: The novel "baseline-referenced Delta radiomics" model demonstrates that by establishing a baseline reference function that captures the pre-treatment radiomic feature-tumor size relationship, one can effectively assess post-treatment changes even without direct pre-treatment scans for every patient.
  • Post-Treatment Scans Alone Can Be Powerful: A significant practical advantage of the new model is its ability to predict response using only post-treatment scans, overcoming a major limitation of traditional Delta radiomics.
  • Delta Radiomics Outperforms Pretreatment: As expected, models incorporating post-treatment changes (Delta radiomics and baseline-referenced Delta radiomics) significantly outperformed models relying solely on pretreatment radiomics.
  • Specific Texture Features Matter: Features like GLRLM_RLNU, NGLDM_Coarseness, and GLZLM_GLNU emerged as top predictors, suggesting that changes in tumor texture are critical indicators of treatment response.

Key Results

The baseline-referenced Delta radiomics model performed comparably or better than established radiomics models in predicting tumor response in patients with liver metastases.

  • Balanced Accuracy: Ranged from 80% to 90% across models and cancer types.
  • Sensitivity and Specificity: Achieved high values, ranging from 0.66 to 0.97 and 0.81 to 0.97, respectively.
  • ROC AUC: In CRC, the AUC for Delta radiomics was 0.82 ± 0.08, and for baseline-referenced Delta radiomics, it was 0.93 ± 0.04. In BC, the AUCs were 0.90 ± 0.04 and 0.91 ± 0.04, respectively, showing substantial improvement over pretreatment radiomics.
  • Model Comparison: The baseline-referenced Delta radiomics model consistently showed higher correlations than the Delta radiomic model, particularly in the CRC cohort.

Limitations and Future Work

The authors acknowledge several limitations:

  • Retrospective Study: The study design limits the ability to control for all confounding variables.
  • Limited Sample Size and Heterogeneous Treatments: Especially in the BC cohort, the variability in treatment regimens might have influenced the results and generalizability.
  • CT Limitations: CT scans have inherent limitations in fully characterizing all tumor features; future work could incorporate MRI and PET.
  • RECIST 1.1 Limitations: The reliance on RECIST 1.1 for response assessment has known limitations.
  • Imaging Heterogeneity: Data from different centers and scanners were used, though harmonization was applied.
  • No Imaging Filters: The study did not employ filters like Gaussian or Wavelets, which could be explored in future work.

Future directions suggested include:

  • Prospective Validation: Confirming these findings in larger, prospective cohorts.
  • Longitudinal Studies: Applying the model to predict long-term outcomes and response durability.
  • Integration with Other Modalities: Extending the baseline-referenced Delta radiomics concept to MRI and PET.
  • Radiogenomics: Identifying associations between imaging features and molecular characteristics.
  • Data Augmentation: Exploring data augmentation techniques and optimal feature selection.

Summary updated

Abstract

<b>Background/Objectives</b>: Radiomic features exhibit a correlation with tumor size on pretreatment images. However, on post-treatment images, this association is influenced by treatment efficacy and varies between responders and non-responders. This study introduces a novel model, called baseline-referenced Delta radiomics, which integrates the association between radiomic features and tumor size into Delta radiomics to predict chemotherapy response in liver metastases from breast cancer (BC) and colorectal cancer (CRC). <b>Materials and Methods</b>: A retrospective study analyzed contrast-enhanced computed tomography (CT) scans of 83 BC patients and 84 CRC patients. Among these, 57 BC patients with 106 liver lesions and 37 CRC patients with 109 lesions underwent post-treatment imaging after systemic chemotherapy. Radiomic features were extracted from up to three lesions per patient following manual segmentation. Tumor response was assessed by measuring the longest diameter and classified according to RECIST 1.1 criteria as progressive disease (PD), partial response (PR), or stable disease (SD). Classification models were developed to predict chemotherapy response using pretreatment data only, Delta radiomics, and baseline-referenced Delta radiomics. Model performance was evaluated using confusion matrix metrics. <b>Results</b>: Baseline-referenced Delta radiomics performed comparably or better than established radiomics models in predicting tumor response in chemotherapy-treated patients with liver metastases. The sensitivity, specificity, and balanced accuracy in predicting response ranged from 0.66 to 0.97, 0.81 to 0.97, and 80% to 90%, respectively. <b>Conclusions</b>: By integrating the relationship between radiomic features and tumor size into Delta radiomics, baseline-referenced Delta radiomics offers a promising approach for predicting chemotherapy response in liver metastases from breast and colorectal cancer.
BibTeX
@inproceedings{gennaro2025delta,
  author = {Nicolò Gennaro and Moataz Soliman and Amir A. Borhani and Linda C. Kelahan and Hatice Savas and Ryan Avery and Kamal Subedi and Tugce Agirlar Trabzonlu and Chase Krumpelman and Vahid Yaghmai and Young Kwang Chae and Jochen H. Lorch and Devalingam Mahalingam and Mary F. Mulcahy and Al B. Benson and Ulaş Bağcı and Yury Velichko},
  title = {{Delta Radiomics and Tumor Size: A New Predictive Radiomics Model for Chemotherapy Response in Liver Metastases from Breast and Colorectal Cancer}},
  booktitle = {Tomography},
  year = {2025},
  doi = {10.3390/tomography11030020},
}
Figure