Delta Radiomics in Head and Neck Cancers

Version	Summary	Created by	Modification	Content Size	Created at	Operation
1		Camil Mirestean	--	2357	2023-06-30 09:42:26	\|
2	format change	Peter Tang	+ 1 word(s)	2358	2023-06-30 09:55:22	\|

This entry is adapted from the peer-reviewed paper 10.3390/diagnostics13122045

Delta (Δ) radiomics, a concept based on the variation of parameters extracted from medical imaging using artificial intelligence (AI) algorithms, demonstrates its potential as a predictive biomarker of treatment response in head and neck cancers (HNC). The concept of image-guided radiotherapy (IGRT), including computer tomography simulation (CT) and position control imaging with cone-beam-computed tomography (CBCT), now offers new perspectives for radiomics applied in radiotherapy. The use of Δ features of texture, shape, and size, both from the primary tumor and from the tumor-involved lymph nodes, demonstrates the best predictive accuracy. If, in the case of treatment response, promising Δ radiomics results could be obtained, even after 24 h from the start of treatment, for radiation-induced xerostomia, the evaluation of Δ radiomics in the middle of treatment could be recommended.

radiomics head and neck cancers CBCT radiotherapy biomarker predictive CT radio-chemotherapy

1. Introduction

Concurrent chemo-radiotherapy (CCRT) as a single treatment, or induction chemotherapy (IC) followed by CCRT, are part of the approach strategies for locally advanced head and neck squamous cell carcinoma (LA-HNSCC). The identification of a biomarker predictive of response could provide a solution for treatment stratification, in the context of high recurrence rates, especially for those associated with loco-regional recurrence. The proposed concept of radiomics involves extracting a large volume of data from medical imaging to improve diagnostic accuracy, but also to create prognostic and predictive models. The researchers propose to identify arguments for the use of a new biomarker (variation Δ of radiomics features during treatment) predictive of the response of head and neck cancers to multimodal treatment by chemo-radiotherapy. The guiding of treatment based on a biomarker would have as a consequence the implementation of de-escalation strategies in order to limit toxicities or to escalate the treatment dose to increase the treatment response rate. Likewise, the prediction, by radiomics models, of the risk of toxicity, such as xerostomia, would have the consequence of replanning in order to limit doses to certain radiosensitive structures ^[1]^[2]^[3].

2. Radiomics—An Emerging Role in HNC

In the last decade, the development of image analysis processes and the rapid increase in the volume of high-quality medical data have facilitated the extraction of quantitative features from high-resolution medical imaging. The transformation of these images into data, invisible to the eye of the examiner, was called radiomics by Gillies and colleagues. The authors particularly note the potential of the method to facilitate clinical decisions and to improve the management of cancer patients. Even if radiomics is frequently used in clinical research and especially in oncology, the value of an analysis of feature variation (Δ-radiomics) during treatment should not be neglected, generally having pre-treatment and post-treatment images as references. If, in the early stages, the therapeutic success rate is relatively high, for LA-HNSCC cases treated by combining radiotherapy with Cisplatin or Anti-EGFR (epidermal growth factor receptor) Cetuximab, 40% of cases will not respond to treatment. After 2–3 years, 50–60% will present loco-regional recurrence, but even the rates of distant failure through metastasis of 20–30%, or 2–4% per year, of the second primary cannot be ignored ^[1]^[2]^[3]^[4].

Radiomics has two major advantages: it could be more specific, but it is also non-invasive. As a marker of tumor heterogeneity, radiomics has an advantage over biopsy due to the possibility of evaluating the tumor in toto. Whether in computer tomography (CT), magnetic resonance imaging (MRI), positron emission tomography (PET) or even ultrasonography (US), mammography, or digital radiographs (XRD), medical images are analyzed and extracted radiomic features could be used to build diagnostic, predictive or prognostic models ^[1]^[5].

Radiomics is therefore the ideal answer for a partnership between the concept of precision medicine and, in particular, of precision oncology and non-invasive biomarkers that could provide a digital tumor phenotype. Using mathematical algorithms, radiomics is based on certain steps, including the initial selection of images, the delimitation of the region of interest (ROI) for two-dimensional images or of the volumes of interest (VOI) for three-dimensional images selected manually, semi-automatically or automatically. Image processing and feature extraction, image analysis and the construction of a model are also essential steps of radiomics. Jha et al. consider that the digital tumor phenotype offered by radiomics will play a decisive role in precision oncology in the future. Radiomics could provide essential data not only about tumor heterogeneity, but also about the tumor microenvironment (TME) ^[6]^[7].

The same applications as in the case of radiomics, including differential diagnosis, prognosis biomarker, prediction of treatment response, but also those anticipating the risk of toxicities associated with the treatment, can be translated to the Δ radiomics concept. Based on recordings of tumor heterogeneity before and after treatment, Δ radiomics analysis could offer a new perspective for chemotherapy and radiotherapy response prediction in HNC. A systematic review, that aimed to identify Δ radiomics studies from Embase, PubMed, and ScienceDirect databases, identified forty-eight studies that corresponded to the selection criteria, of which six studies, representing 12.5%, are related to HNC. It should be noted that the large number of Δ radiomics studies is equal to the number of studies that analyze the radiomic feature dynamics in gastrointestinal cancers (other than rectal cancer) and in rectal cancer. Only in the case of lung cancer were a doubled number of Δ radiomics analysis studies identified compared to HNC ^[2]^[6]^[8]^[9].

3. Δ Radiomics in HNC—From Image-Guided Radiotherapy (IGRT) and Adaptive Radiotherapy (ART) to Treatment Response/Toxicity Prediction

Tran et al. propose a non-invasive method, based on Δ radiomics, of quantitative ultrasound (QUS) extracted from metastatic lymph nodes in order to predict the therapeutic response in the first 24 h after the start of treatment with radical radiotherapy. Spectral and texture features extracted from QUS, at the beginning of radiotherapy and at 24 h, 1 week, and 4 weeks after the beginning of treatment, were evaluated in the study. By dividing the patients according to the response to treatment at 3 months into partial and complete responders, the authors build a predictive Δ radiomic model. A single-feature naïve-Bayes classification provides the best correlation with the therapeutic response. The predictive accuracies were 80, 86, and 85% at 24 h, 1 week, and 4 weeks, respectively. The authors note that, even if QUS Δ radiomics, evaluated at 1 week and 4 weeks, offers better results, the method also offers reasonable accuracy for early radiotherapy treatment response prediction ^[10].

Dividing a cohort of 93 HNC patients into a training (60 cases) and validation set (33 cases), Sellami and colleagues evaluated, in a Δ radiomic study, 88 features extracted from the gross tumor volume (GTV) delineated on CBCT images. After the selection by receiver operating characteristic (ROC) curves of those radiomic features significant for the weekly response to CCRT, those radiomic features that had a significant variation in at least 5% of the cases were evaluated. After these repeated exclusions of redundant radiomic features, only Coarseness was considered as a Δ radiomic feature predictor of response to chemo-radiotherapy. The association of clinical variables with the model based on Δ-radiomics variation could increase the accuracy of treatment response prediction. Thus, a combined model that included the hemoglobin value demonstrated predictive superiority over the Δ radiomic model ^[11].

Adaptive radiotherapy (ART) originated at the end of the 1990s, but Yan et al. highlight the concept for the first time in an article published in 2010. The National Radiotherapy Advisory Group in the UK has been recommending four-dimensional adaptive radiotherapy (4D-ART) as the future standard in radiotherapy since 2007, it being considered an “extremely encouraging and greatly promoting adaptive radiotherapy”. ART could contribute to improving radiotherapy patient outcomes and methods such as auto-segmentation, automated planning, and deformable registration algorithms are proposed as a response to anatomical and functional variations of the structures involved in treatment planning. HNC, lung, and cervix cancers are mentioned as potential beneficiaries of ART. Currently, MRI-guided online adaptive radiotherapy is considered as a method with possible major benefits for liver, pancreas, abdominal lymph nodes, and prostate cancers’ treatment planning. Weight loss and tumor response are considered essential factors associated with the need to correct the treatment plans. Avgousti et al. mentions, in an analysis that includes 85 articles, four major criteria advocating for ART: >10% weight loss; >1 cm deviation in external contour; >5% variations compared to the accepted doses; and variations of the planned doses in the periphery of the target volumes. The authors mention the value of ART for HNC radiotherapy and the ability of the method to decrease toxicity and improve local control. Hybrid adaptation, including image guidance combined with offline re-planning, is considered the optimal option to respond to organ motion and tumor shrinkage. Using fluoro-2-deoxyglucose (FGD) positron emission tomography (PET)-CT imaging, before and after CCRT treatment but also weekly after each chemotherapy session, for HNC cases, Yan and colleagues conceptually propose adaptive dose painting by number (DPbN) based on the voxel dose response matrix obtained by evaluating the voxel dose response matrix and the baseline standardized uptake value (SUV). The conclusions mention the existence of tumor voxels not correlated with baseline SUV as the main regions of radio-resistance. For HNC not related to human papillomavirus (HPV), radio-resistance tumor subvolumes are identified requiring >100 Gy in order to obtain adequate local control. DPbN is considered feasible as an adapted method to radiobiological heterogeneity in order to overcome radio-resistance ^[12]^[13]^[14]^[15]^[16]^[17].

The early prediction of the anatomical variations of the structures’ volume involved in HNC radiotherapy treatment could provide an indication of ART re-planning to avoid tumor coverage errors and radiation overdose of OARs. In order to identify those cases associated with major anatomical variation as a result of the early response to irradiation, a study based on the weekly assessment of CBCT used for IGRT was proposed. A total of 104 Δ-radiomics were extracted from the clinical target volume (CTV) and parotid gland volume. A model composed of 13 features extracted from CTV and 6 features extracted from parotid glands offered a 0.90 accuracy, 0.95 sensitivity, and 0.86 specificity performance for the previously mentioned end-point. The variation of three features, including the gray level matrix features family, were identified as significant in both analyzed structures ^[18].

An ability evaluation of radiomic features kinetics, extracted from daily mega-voltage CT (MVCT) scans for HNC radiotherapy IGRT used to predict moderate to severe xerostomia, was compared to the dose/volume parameters. The variation Δ of radiomic features extracted from MVCT for the contralateral parotid gland were evaluated as a predictor of moderate to severe xerostomia at 6, 12, and 24 months after treatment. The Δ radiomic model was compared with a pre-treatment dosimetric model and with a mixed dosimetric–radiomic model. For the models internal validation, cross-validation and bootstrapping were used. The Δ radiomic performances were quantified both on training sets and on test sets; area under the curve (AUC) values of 46%, 33%, and 26% for the moderate to severe xerostomia predictive rate at 6, 12, and 24 months, respectively, were obtained by testing the model. Radiomic Δ variations obtained from MVCT, comparing the baseline treatment’s half features, demonstrated superior predictive power compared to the pre-treatment dosimetric models. Pota et al. also mentions the use of the radiomic parameters’ dynamics, extracted from CT images until the halfway point of the treatment, as early predictors of radiotherapy-induced parotid shrinkage, but also for treatment-related toxicity. The fuzzy logic concept, a method of grouping data with the same characteristics into fuzzy sets and a model with the capacity of managing uncertain data and building rule-based classifiers, was proposed by the authors. Weekly CT changes during treatment quantified in Δ radiomic features were assessed on 68 treatment planning CTs and 340 weekly control CTs as a predictor of moderate to severe xerostomia 12 months after radiotherapy. Geometric, intensity, and texture features were evaluated for their predictive power in relation to the reference models based on mean radiation dose for contralateral parotid gland and nadir xerostomia score (Xerbaseline). The Δ variation of the contralateral parotid gland surface was the most predictive feature correlated with moderate and severe xerostomia at 12 months. The Δ changes of the surface of the parotid gland, 3 weeks after the treatment started, demonstrated superior predictive capacity compared to the pre-treatment models. QUS Δ radiomics obtained during CCRT is also proposed as a predictor of recurrence in HNSCC. A total of 31 spectral and texture radiomic features were extracted from a tumor-involved lymph node. The study group included 51 cases treated with a total radiotherapy dose of 70 Gy in 33 daily fractions of IMRT. The radiomic data were extracted before the treatment, and at one and four weeks after the beginning of the treatment. The study included patients with cancers of the oropharynx, larynx, hypopharynx, and of unknown primary; all cases were treated with platinum-based chemotherapy or Cetuximab. In a median follow-up of 38 months (range 7–64 months), relapse was identified in 17 cases. The model predicted the recurrence with an accuracy of 80% and 82% using Δ radiomics at one week and 4 weeks, respectively. The necessity of anticipating the result of chemo-radiotherapy treatment in HNC is also mentioned by Morgan et al., which also highlights the advantage of a response-adapted concept in order to escalate, de-escalate, or modify the treatment. In a retrospective analysis, the authors analyze, in a single institution study, the variation of radiomic features between baseline CT and CBCT, recorded at 1 day and 21 days during radiotherapy treatment. The analysis includes the primary tumor and metastases lymph nodes, with both local or lymph node failure being analyzed.

Evaluating the risk of acute xerostomia in a 35 patient lot with stage I-IVB nasopharyngeal cancer treated with IMRT radiotherapy, Liu et al. compared the model based on the variation Δ of CT radiomic features extracted from the parotid glands and a model based on saliva amount quantification. Radiomic features were extracted at fractions 0, 10, 20, and 30. Four patients were subsequently added to the lot for independent testing of the model. The delineation of the parotid gland was performed manually by the radiation oncologist, and the saliva was collected every 10 days during the treatment. Both the amount of saliva and the Radiation Therapy Oncology Group (RTOG) acute xerostomia score were used to quantify acute xerostomia. A total of 1703 radiomic features were extracted from each CT image of the parotid gland. RidgeCV and recursive feature elimination (RFE) methods were used for features selection and feature matrix reduction. Totals of 8 baseline radiomic features and 14 radiomic features extracted after 10 days of treatment were identified as significant for predicting acute xerostomia. The model that compared the variation of radiomic features between the baseline value and the value recorded at 10 treatment fractions offered the best predictive power, with a precision of 0.9220 and a sensitivity of 100% compared to the model based on saliva amount ^[19].

References

Gillies, R.J.; Kinahan, P.E.; Hricak, H. Radiomics: Images Are More than Pictures, They Are Data. Radiology 2016, 278, 563–577.
Nardone, V.; Reginelli, A.; Grassi, R.; Boldrini, L.; Vacca, G.; D’Ippolito, E.; Annunziata, S.; Farchione, A.; Belfiore, M.P.; Desideri, I.; et al. Delta radiomics: A systematic review. Radiol. Med. 2021, 126, 1571–1583.
Denaro, N.; Merlano, M.C.; Russi, E.G. Follow-up in Head and Neck Cancer: Do More Does It Mean Do Better? A Systematic Review and Our Proposal Based on Our Experience. Clin. Exp. Otorhinolaryngol. 2016, 9, 287–297.
Sturgis, E.M.; Miller, R.H. Second primary malignancies in the head and neck cancer patient. Ann. Otol. Rhinol. Laryngol. 1995, 104, 946–954.
Fave, X.; Zhang, L.; Yang, J.; Mackin, D.; Balter, P.; Gomez, D.; Followill, D.; Jones, A.K.; Stingo, F.; Liao, Z.; et al. Delta-radiomics features for the prediction of patient outcomes in non–small cell lung cancer. Sci. Rep. 2017, 7, 588.
Jha, A.K.; Mithun, S.; Purandare, N.C.; Kumar, R.; Rangarajan, V.; Wee, L.; Dekker, A. Radiomics: A quantitative imaging biomarker in precision oncology. Nucl. Med. Commun. 2022, 43, 483–493.
Wang, X.; Xie, T.; Luo, J.; Zhou, Z.; Yu, X.; Guo, X. Radiomics predicts the prognosis of patients with locally advanced breast cancer by reflecting the heterogeneity of tumor cells and the tumor microenvironment. Breast Cancer Res. 2022, 24, 20.
Iancu, R.I.; Zara, A.D.; Mirestean, C.C.; Iancu, D.P.T. Radiomics in Head and Neck Cancers Radiotherapy. Promises and Challenges. Maedica (Bucur) 2021, 16, 482–488.
Mirestean, C.C.; Pagute, O.; Buzea, C.; Iancu, R.I.; Iancu, D.T. Radiomic Machine Learning and Texture Analysis—New Horizons for Head and Neck Oncology. Maedica (Bucur) 2019, 14, 126–130.
Tran, W.T.; Suraweera, H.; Quiaoit, K.; DiCenzo, D.; Fatima, K.; Jang, D.; Bhardwaj, D.; Kolios, C.; Karam, I.; Poon, I.; et al. Quantitative ultrasound delta-radiomics during radiotherapy for monitoring treatment responses in head and neck malignancies. Future Sci. OA 2020, 6, FSO624.
Sellami, S.; Bourbonne, V.; Hatt, M.; Tixier, F.; Bouzid, D.; Lucia, F.; Pradier, O.; Goasduff, G.; Visvikis, D.; Schick, U. Predicting response to radiotherapy of head and neck squamous cell carcinoma using radiomics from cone-beam CT images. Acta Oncol. 2022, 61, 73–80.
Yan, D. Adaptive radiotherapy: Merging principle into clinical practice. Semin. Radiat. Oncol. 2010, 20, 79–83.
National Radiotherapy Advisory Group. Radiotherapy: Developing a World Class Service for England. 2007. Available online: https://www.axrem.org.uk/radiotheraphy_papers/DH_Radiotheraphy_developing_first_class_service_NRAG.pdf (accessed on 3 March 2023).
Nierer, L.; Eze, C.; da Silva Mendes, V.; Braun, J.; Thum, P.; von Bestenbostel, R.; Kurz, C.; Landry, G.; Reiner, M.; Niyazi, M.; et al. Dosimetric benefit of MR-guided online adaptive radiotherapy in different tumor entities: Liver, lung, abdominal lymph nodes, pancreas and prostate. Radiat. Oncol. 2022, 17, 53.
Avgousti, R.; Antypas, C.; Armpilia, C.; Simopoulou, F.; Liakouli, Z.; Karaiskos, P.; Kouloulias, V.; Kyrodimos, E.; Moulopoulos, L.A.; Zygogianni, A. Adaptive radiation therapy: When, how and what are the benefits that literature provides? Cancer Radiother. 2022, 26, 622–636.
Oh, S.; Stewart, J.; Moseley, J.; Kelly, V.; Lim, K.; Xie, J.; Fyles, A.; Brock, K.K.; Lundin, A.; Rehbinder, H.; et al. Hybrid adaptive radiotherapy with on-line MRI in cervix cancer IMRT. Radiother. Oncol. 2014, 110, 323–328.
Yan, D.; Chen, S.; Krauss, D.J.; Chen, P.Y.; Chinnaiyan, P.; Wilson, G.D. Tumor Voxel Dose-Response Matrix and Dose Prescription Function Derived Using 18F-FDG PET/CT Images for Adaptive Dose Painting by Number. Int. J. Radiat. Oncol. Biol. Phys. 2019, 104, 207–218.
Iliadou, V.; Kakkos, I.; Karaiskos, P.; Kouloulias, V.; Platoni, K.; Zygogianni, A.; Matsopoulos, G.K. Early Prediction of Planning Adaptation Requirement Indication Due to Volumetric Alterations in Head and Neck Cancer Radiotherapy: A Machine Learning Approach. Cancers 2022, 14, 3573.
Liu, Y.; Shi, H.; Huang, S.; Chen, X.; Zhou, H.; Chang, H.; Xia, Y.; Wang, G.; Yang, X. Early prediction of acute xerostomia during radiation therapy for nasopharyngeal cancer based on delta radiomics from CT images. Quant. Imaging Med. Surg. 2019, 9, 1288–1302.

© Text is available under the terms and conditions of the Creative Commons Attribution (CC BY) license; additional terms may apply. By using this site, you agree to the Terms and Conditions and Privacy Policy.

Upload a video for this entry

Information

Subjects: Oncology

Contributors MDPI registered users' name will be linked to their SciProfiles pages. To register with us, please refer to https://encyclopedia.pub/register :

Camil Ciprian Mireștean

Roxana Irina Iancu

Dragoș Petru Teodor Iancu

View Times: 366

Update Date: 30 Jun 2023

Table of Contents

Video Upload Options

Confirm

1. Introduction

2. Radiomics—An Emerging Role in HNC

3. Δ Radiomics in HNC—From Image-Guided Radiotherapy (IGRT) and Adaptive Radiotherapy (ART) to Treatment Response/Toxicity Prediction

References