Brachytherapy outcome modeling in cervical cancer patients: A predictive machine learning study on patient-specific clinical, physical and dosimetric parameters
Introduction
Cervical cancer is the fourth reason for death among women worldwide (1). The globally established curative setting for locally advanced (LACC) patients, that is, patients with the International Federation of Gynecology and Obstetrics (FIGO) staging system, Stages IB−IVA, include brachytherapy (BT) consecutive external beam radio-chemotherapy (2). Several trials revealed that BT is a vital part of treatment as a dose-escalation approach. It is also declared that doses of more than 80−85 Gy to the high-risk clinical target volume (HR-CTV) achievable by BT can significantly improve outcomes in terms of local control (3,4). Furthermore, during the last decades, BT has experienced technological revolutions by using three-dimensional images, including computed tomography (CT) and magnetic resonance imaging (MRI), to deliver prescribed doses to target volumes instead of historical reference point A (5). Moreover, image-guided BT (IGBT) benefits from the most proper and efficient applicators used to transfer high-dose-rate (HDR) radioactive sources into the patient body and very close to or inside the tumor.
Comprehensive studies have identified that several clinical, hematological, pathological, and treatment-related variables significantly impact LACC patient outcomes (6). For instance, Lee et al. (7) introduced posttreatment neutrophil to lymphocyte ratio (NLR) as a prognostic factor. It is also revealed that some tumor-specific parameters such as size, stage, and squamous cell carcinoma antigen (SCC Ag) significantly change the patients’ outcomes ([8], [9], [10], [11]). On the other hand, treatment-related and dosimetric factors such as dose per fraction and dose-volume histogram (DVH) parameters are introduced as prognostic measures (12,13). Although the relationship between external beam radiotherapy (EBRT) and concurrent chemotherapy parameters with the outcome of LACC patients was investigated comprehensively, there is a lack of such research specific to BT parameters as a different treatment technique in terms of radiobiological, physical, and dosimetric perspectives (14,15).
Predictive modeling is an advanced analytical approach to personalizing treatment by mining features extracted from daily clinical practice data. A wide range of research has been made in this era by advances in machine and deep learning algorithms and understanding their feasible roles in predictive modeling (16). Studies have revealed that features extracted from diagnostic and therapeutic data combined with machine learning algorithms could predict patients' survival, tumor fate, healthy organ toxicities, and recurrence after different kinds of therapy, including radiotherapy for almost all cancers undergoing treatments (17,18). These models are new ways to tailor the therapy and make the best clinical decisions. Several studies have applied machine learning models to learn the patterns in cervical cancer data for cancer prediction, classification, outcome modeling, and therapy-associated healthy tissue toxicity prediction ([19], [20], [21], [22], [23]).
It is revealed that despite standard treatment of LACC, about 20−30% of patients experience a recurrence 2 years after treatment. For this issue, accurate outcome prediction would be feasible for high-risk patients who may benefit from other adjuvant treatments like chemotherapy or surgery after BT. Although several studies have been published for LACC outcome modeling, there is a lack of knowledge on BT-only predictive modeling. There are many parameters in BT that may be predictive. For example, the predictive value of some insertion criteria based on two-dimensional radiographs was investigated by Katz et al. They found a correlation between standard intra-cavitary (IC) insertion and central disease control (24,25).
The present research aimed to assess cervical cancer brachytherapy outcome modeling using patient-specific applicator insertion indexes, BT planning measures, including physical, dosimetric, radiobiological parameters, and individualized clinical factors. Our approach recruited several machine learning algorithms and statistical tests for model development and comparison. The novelties of this work are the quantification of applicator insertion geometry based on three-dimensional images (MRI and CT) and comprehensive parameter extraction from physical, dosimetric, and radiobiological aspects of cervical cancer patients HDR-BT and then using them for outcome model creation.
Section snippets
Methods and material
For conducting this research study, all ethical issues were confirmed by the local ethics committee of the Iran University of Medical Sciences, with ethics code number 1399.583. Before starting the research, informed consent was obtained from all patients enrolled in this study.
Patients
In total, 111 LACC patients with Stage IB−IVA that had finished EBRT followed by concurrent cisplatin-based chemotherapy were analyzed in this study. Table 1 summarized characteristics of patients. The average duration of followup was 8 months, and at the end of the study, there were 42 (38%) nonresponders and 69 responders (62%) between eligible patients.
Our results on univariate feature analysis are shown in Table 3. Each model depicted ten high-rank features with their odds ratio (OR).
Discussion
Dose escalation using brachytherapy without exceeding tolerance doses of organs at risk is the main advantage of this treatment modality in cervical cancer treatment (48). However, studies identified that about 30% of patients experience disease recurrence (49). Furthermore, it is debated that the diversity of BT-related factors may affect treatment outcomes (50,51). In this era, finding highly relevant measures in terms of outcome predictive biomarkers is still an ongoing research solution.
Conclusion
The present study showed that parameters extracted from brachytherapy planning, including source-related parameters, dosimetric factors, and clinical measures, alone or in combination, could predict the clinical response of locally advanced cervical cancer. It was observed that machine learning algorithms are feasible methods for model development. It was also identified that combining physical and dosimetric parameters and the Random Forest machine learning algorithm would result in the
References (70)
- et al.
Cervical cancer worldwide
Curr Probl Cancer
(2018) - et al.
Curative radiation therapy for locally advanced cervical cancer: brachytherapy is NOT optional
Int J Radiat Oncol Biol Phys
(2014) - et al.
Image guided brachytherapy in locally advanced cervical cancer: improved pelvic control and survival in RetroEMBRACE, a multicenter cohort study
Radiother Oncol
(2016) - et al.
Effect of tumor dose, volume and overall treatment time on local control after radiochemotherapy including MRI guided brachytherapy of locally advanced cervical cancer
Radiather Oncol
(2016) - et al.
Knowledge-based dose prediction models to inform gynecologic brachytherapy needle supplementation for locally advanced cervical cancer
Brachytherapy
(2021) - et al.
Clinical outcome of protocol based image (MRI) guided adaptive brachytherapy combined with 3D conformal radiotherapy with or without chemotherapy in patients with locally advanced cervical cancer
Radiother Oncol
(2011) - et al.
The role of squamous cell carcinoma antigen (SCC Ag) in outcome prediction after concurrent chemoradiotherapy and treatment decisions for patients with cervical cancer
Radiat Oncol
(2019) - et al.
Tumor diameter and volume assessed by magnetic resonance imaging in the prediction of outcome for invasive cervical cancer
Gynecol Oncol
(2001) - et al.
Dose–volume histogram parameters and local tumor control in magnetic resonance image–guided cervical cancer brachytherapy
Int J Radiat Oncol Biol Phys
(2009) - et al.
Risk assessment model for overall survival in patients with locally advanced cervical cancer treated with definitive concurrent chemoradiotherapy
Gynecol Oncol
(2013)
MRI radiomic analysis of IMRT-induced bladder wall changes in prostate cancer patients: a relationship with radiation dose and toxicity
J Med Imaging Radiat Sci
Machine learning for assisting cervical cancer diagnosis: an ensemble approach
Future Gener Comput Syst
A fully-automated deep learning pipeline for cervical cancer classification
Expert Syst Appl
Quantification of intracavitary brachytherapy parameters and correlation with outcome in patients with carcinoma of the cervix
Int J Radiat Oncol Biol Phys
Recommendations from Gynaecological (GYN) GEC-ESTRO Working Group (IV): basic principles and parameters for mr imaging within the frame of image based adaptive cervix cancer brachytherapy
Radiather Oncol
Recommendations from Gynaecological (GYN) GEC-ESTRO Working Group☆(I): concepts and terms in 3D image based 3D treatment planning in cervix cancer brachytherapy with emphasis on MRI assessment of GTV and CTV
Radiother Oncol
Recommendations from gynaecological (GYN) GEC ESTRO working group (II): concepts and terms in 3D image-based treatment planning in cervix cancer brachytherapy—3D dose volume parameters and aspects of 3D image-based anatomy, radiation physics, radiobiology
Radiother Oncol
American Brachytherapy Society consensus guidelines for locally advanced carcinoma of the cervix. Part II: high-dose-rate brachytherapy
Brachytherapy
The role of biologically effective dose (BED) in clinical oncology
Clin Oncol
A simple method of obtaining equivalent doses for use in HDR brachytherapy
Int J Radiat Oncol Biol Phys
Role of conventional and diffusion weighted MRI in predicting treatment response after low dose radiation and chemotherapy in locally advanced carcinoma cervix
Radiother Oncol
Hybrid (CT/MRI based) vs. MRI only based image-guided brachytherapy in cervical cancer: dosimetry comparisons and clinical outcome
Brachytherapy
Physics contributions and clinical outcome with 3D-MRI–based pulsed-dose-rate intracavitary brachytherapy in cervical cancer patients
Int J Radiat Oncol Biol Phys
Use of functional magnetic resonance imaging in cervical cancer patients with incomplete response on positron emission tomography/computed tomography after image-based high-dose-rate brachytherapy
Int J Radiat Oncol Biol Phys
Differentiation of recurrent glioblastoma from radiation necrosis using diffusion radiomics with machine learning model development and external validation
Sci Rep
A systematic analysis of performance measures for classification tasks
Inf Process Manag
Tumor shrinkage during chemoradiation in locally advanced cervical cancer patients: prognostic significance, and impact for image-guided adaptive brachytherapy
Int J Radiat Oncol Biol Phys
Long-term effectiveness and safety of image-based, transperineal combined intracavitary and interstitial brachytherapy in treatment of locally advanced cervical cancer
Brachytherapy
Clinical outcome of recurrent locally advanced cervical cancer (LACC) submitted to primary multimodality therapies
Gynecol Oncol
High-dose-rate brachytherapy boost for locally advanced cervical cancer: Oncological outcome and toxicity analysis of 4 fractionation schemes
Clin Transl Radiat Oncol
The American Brachytherapy Society recommendations for high-dose-rate brachytherapy for carcinoma of the cervix
Int J Radiat Oncol Biol Phys
Cochlea CT radiomics predicts chemoradiotherapy induced sensorineural hearing loss in head and neck cancer patients: a machine learning and multi-variable modelling study
Physica Medica
Magnetic resonance image guided adaptive brachytherapy in locally advanced cervical cancer: an experience from a tertiary cancer center in a low and middle income countries setting
Int J Radiat Oncol Biol Phys
Prognostic implications of uterine cervical cancer regression during chemoradiation evaluated by the T-Score in the multicenter EMBRACE I Study
Int J Radiat Oncol Biol Phys
Big data and machine learning in radiation oncology: state of the art and future prospects
Cancer Lett
Cited by (2)
Disclosures: This research was supported by grant No. [16525] from the School of Medicine, Iran University of Medical Sciences (IUMS). There is no conflict of interest declared in this article.