1. Introduction
Cervical cancer is the fourth most commonly diagnosed cancer and the fourth leading cause of mortality among women worldwide. There were over 570,000 new cases annually and over 311,000 deaths in 2018 globally [1]. More than two-thirds of the patients are diagnosed with locally advanced cervical cancer (LACC) [2]. Concurrent chemoradiotherapy (CCRT) can lead to good prognosis for LACC [3], and the brachytherapy as a part of standard treatment regimen has further improved local control and overall survival, but the chemoradiotherapy resistance and local relapse also occur among 40% of LACC patients [4, 5]. Early treatment response correlates to tumor response and survival [6]. Hematologic toxicity is the most common acute side effect of CCRT, which affect treatment efficacy and sometimes can be life-threatening [7]. Pretreatment prediction of early tumor response, survival, and hematologic toxicities may assist physicians to adjust radiotherapy planning and further improve the quality of life.
Previous studies demonstrated that the predictive efficacy of clinicopathological model for treatment outcomes is limited in LACC [8, 9]. The therapeutic effect can be markedly different even in the same clinical stage. Currently, radiomics, a technique that can extract high-throughput quantitative features from imaging such as CT, PET, and MRI, has emerged as a promising tool for the assessment of treatment outcomes [10, 11]. Radiomics can reflect the tumor intrinsic properties and can be used as independent predictors of survival outcome with higher predictive ability than traditional clinical parameters alone [12, 13].
Radiomics has been found to be related to lymph node metastases [14, 15], treatment response [16-18], patient survival [19], and tumor recurrence [16, 20] for cervical cancer patients [21]. So far as we know, there are no studies to investigate the radiomics predictive ability of CT images from radiotherapy planning system. Additionally, myelosuppression is the most common toxicity in pelvic radiotherapy with the Grade 3 or higher hematological toxicities of more than 30% [22]. Bone marrow is the most radiosensitive pelvic organ, and approximately 40% of the total body BM reserve lies within the pelvic bones, especially pelvis and sacral vertebrae, which always are within the irradiated fields [23]. The individual differences of the radiation-induced hematologic toxicities caused even by the similar treatment regimens. Previously, radiomics was used to quantify lumbar and femoral mineral loss [24], mineral bone density, and osteoblast activity [25]. Treatment planning CT-based radiomic features of pelvis and sacral vertebrae may be able to predict the hematologic toxicities.
Therefore, we retrospectively examined radiotherapy planning CT images to establish radiomic models for predicting tumor complete response (CR), 5-year overall survival (OS) as well as hematologic toxicities (leucopenia and hypohemoglobin) in LACC patients. We also compared the prediction performance of clinicopathological parameters and radiomic features to further construct a combined model, aiming at better prediction of clinical endpoints in LACC.