European Radiology: The use of radiomics in assessing response to whole-brain radiation therapy

Brain metastases are a serious manifestation of cancer and are directly related to
poor clinical prognosis.
Despite recent clinical advances in radiation oncology, neurosurgery, and systemic therapies, identifying effective treatments for patients with personalized brain metastases remains a challenge
.

Whole-brain radiotherapy (WBRT) irradiates the whole brain and is mainly indicated for patients
with brain metastases who are not candidates for surgery or radiosurgery.
The advantage of WBRT is that it reduces the rate of recurrence and the emergence
of new brain lesions.
In addition, according to clinical trials, patients with brain metastases who respond well to treatment, such as partial response (PR) or complete response (CR), generally have higher overall survival (OS).

However, the objective response rate (ORR) of WBRT is not high, and nearly half of patients do not benefit
from WBRT.
At the same time, the potential benefits of WBRT must be balanced
against radiation-related side effects.
Therefore, it is important to
accurately assess treatment response prior to WBRT.

Currently, primary tumor type and disease-specific graded prognostic assessments (DS-GPAs) are objective prognostic scoring systems for patients with brain metastases and are often used to assist radiation oncologists in determining appropriate WBRT strategies, and rarely to directly assess WBRT treatment response
.
At this stage, radiomics, as an emerging technology, can provide a better understanding
of cancer pathophysiology for clinical decision-making by using a large number of quantitative image features to describe tumor heterogeneity.
Several studies have reported successful use of radiomics to evaluate the response
of Gamma Knife radiosurgery or stereotactic radiosurgery (SRS) therapy in patients with brain metastases.
However, the differential effect of radiomics methods on WBRT treatment response remains unclear
.

In fact, many radiomics models have been developed and have yielded good results
with limited datasets or applications.
However, due to unclear internal mechanisms, interpretability largely hinders the widespread application
of radiomics models.
SHAP is a unified framework for interpreting six different methods for forecasting, defining a class of additive feature importance measures and theoretical results, combined with radiomics to illustrate the model
in an interpretable way.

Recently, a study published in the journal European Radiology constructed a radiomics-clinical machine learning model to evaluate the treatment response of WBRT, and verified the practicality and reproducibility of the model through external validation, providing technical support
for the clinical accurate and effective evaluation of WBRT treatment response.

This study retrospectively recruited 228 patients with brain metastases from two medical centers (184 in the training cohort and 44 in the validation cohort).

According to the neuro-oncology brain metastases response assessment (RANO-BM) criteria, the patient's treatment response is divided into non-response groups and responsive groups
.
For each tumor, 960 features
were extracted from the MRI sequence.
The Minimum Absolute Shrinkage and Selection Operator (LASSO) is used for feature selection
.
Support vector machine (SVM) models are used to construct radiology-clinical models
based on clinical factors and radiological features.
The SHAP method explains the assessed value of SVM models by prioritizing the importance of features.

Three radiological features and three clinical factors were identified to model
them.
In the training and validation cohorts, the radiology-clinical model generated AUCs of 0.
928 (95% CI 0.
901 to 0.
949) and 0.
851 (95% CI 0.
816 to 0.
886), respectively, for assessing treatment response
.
The SHAP summary plot shows the effect of eigenvalues on the model, and the SHAP effort plot shows the integration of
the effects of features on individual responses.

 
Figure ROC curve comparison between training group (A) and validation group (B) to assess WBRT treatment response

The radiology-clinical machine learning model containing CET1-w (3D) MR sequence features proposed in this study can be used to evaluate the treatment response of WBRT, where the SHAP method used can assist clinicians in guiding personalized WBRT strategies
in patients with brain metastases.

Original source:

Yixin Wang,Jinwei Lang,Joey Zhaoyu Zuo,et al.
The radiomic-clinical model using the SHAP method for assessing the treatment response of whole-brain radiotherapy: a multicentric study.
DOI:10.
1007/s00330-022-08887-0