Purpose: FDG-PET is often used in the assessment, treatment planning, and monitoring of locally advanced cervical cancer (LACC), contributing to improved patient care and management. The aim of this study is to examine the utility of FDG-PET radiomics in discriminating between patients with LACC who demonstrated an early response to treatment and those who did not.

Methods: The study uses FDG-PET scans of 21 patients obtained from The Cancer Imaging Archive (TCIA) Cervical Cancer Tumor Heterogeneity (CCTH) collection. Among the study cohort, 12 patients responded to chemoradiation treatment, while the remaining 9 did not. A total of 59 radiomics texture features were extracted from the gross tumor volumes (GTVs). Top radiomics features in terms of differentiating response status were selected using a sequential feature selection (SFS) algorithm. The top two features were used to build a support vector machine (SVM) predictive classifier, with performance assessed by a receiver operating characteristic (ROC) curve analysis.

Results: The top two radiomics features being selected consisted of the gray level co-occurrence matrix (GLCOM)-based cluster prominence and gray level run length matrix (GLRLM)-based length nonuniformity. The SVM predictive classifier built using the top two features had an area under the ROC curve (AUC) of 0.88, showing strong discriminative power between responders and non-responders.

Conclusions: The radiomic features extracted from FDG-PET scans showed a discriminative ability to discern patients with LACC who would have an early response to treatment. This knowledge could have clinical implications for adjusting treatment strategies for patients who might not show an early response.