The Role of PET-Based Radiomic Features in Predicting Local Control of Esophageal Cancer Treated with Concurrent Chemoradiotherapy

Abstract

This study was designed to evaluate the predictive performance of ¹⁸F-fluorodeoxyglucose positron emission tomography (PET)-based radiomic features for local control of esophageal cancer treated with concurrent chemoradiotherapy (CRT). For each of the 30 patients enrolled, 440 radiomic features were extracted from both pre-CRT and mid-CRT PET images. The top 25 features with the highest areas under the receiver operating characteristic curve for identifying local control status were selected as discriminative features. Four machine-learning methods, random forest (RF), support vector machine, logistic regression, and extreme learning machine, were used to build predictive models with clinical features, radiomic features or a combination of both. An RF model incorporating both clinical and radiomic features achieved the best predictive performance, with an accuracy of 93.3%, a specificity of 95.7%, and a sensitivity of 85.7%. Based on risk scores of local failure predicted by this model, the 2-year local control rate and PFS rate were 100.0% (95% CI 100.0-100.0%) and 52.2% (31.8-72.6%) in the low-risk group and 14.3% (0.0-40.2%) and 0.0% (0.0-40.2%) in the high-risk group, respectively. This model may have the potential to stratify patients with different risks of local failure after CRT for esophageal cancer, which may facilitate the delivery of personalized treatment.

Figure 1

The workflow of the development and cross validation of the RF model developed with radiomic features.

Figure 2

Performance of typical discriminative radiomic features for determining the local control of esophageal cancer after CRT. The upper two rows show ROC curves based on features extracted from pre- and mid-CRT SUV images, respectively. The lower two rows show the values of the features extracted from pre- and mid-CRT SUV images plotted against local control status, with 1 on the horizontal axis representing local progression and 0 representing local control, respectively.

Figure 3

PET images of two typical patients with local control (left) and local progression (right), respectively. The skewness_HLL, RP_HLL, cluster promience_HLL, and median_HLL values are 0.164, −1.744, 65280, and −0.038 for the patient with local control and 0.743, −0.278, 24240, and 0.003 for the patient with local progression.

Figure 4

Prediction measures of four models involving specific groups of features. RF, random forest; SVM, support vector machines; LR, logical regression; ELM, extreme learning machine.

Figure 5

Kaplan-Meier plots of the local control and progression-free survival rates in patients with esophageal cancer treated with CRT based on the risk scores derived from the random forest model incorporating both clinical and radiomic features.