1. Introduction

It is estimated that almost 70% of all cancer patients receive radiotherapy during the course of the disease, either for curative or palliative purposes. Modern radiotherapy involves 7 different processes: diagnostic imaging, treatment planning, computer simulation, use of radiotherapy accessories, delivery of radiation treatment, treatment verification and patient follow-up. Anatomical and functional medical imaging has an important role both in the diagnostic stage and in treatment planning, treatment simulation, but also in quality assurance (QA), in the ballistic accuracy verification stage of the delivery of each treatment fraction (1).

Image-guided radiotherapy (IGRT) allows the delivery of a therapeutic dose of radiation to the tumor region while sparing healthy normal tissue. The concept of radiomics, mentioned by Gillies and colleagues describes it as a process which allows the large amounts of data mining in order to create diagnostic, predictive and prognostic models. The authors believe that “images are more than pictures, they are data”, considering that the field of oncology can derive a substantial benefit from radiomics technology. Inclusion of clinical and biological data in radiomics models, associating them with information extracted from images using bioinformatics tools will increase the accuracy and performance of the method and facilitate the implementation of radiomics in clinical practice. Abdollahi et al. analyzed in depth the opportunities and challenges of the new radiomics guided radiotherapy (RGRT) concept, including cancer diagnosis, prognosis, and therapy response evaluation, considering radiomic features “biomarkers towards personalized medicine”. The concept was previously proposed for urgent implementation in routine clinical research in radiotherapy and radiobiology at a national congress of oncology and radiotherapy in 2021 by Mirestean et al. in the lecture “RGRT – We are ready to go!. The RGRT concept involves the use of AI and radiomics to guide physicists and clinicians in order to optimize radiation treatment in cancer. RGRT demonstrates its value in all stages of treatment, from patient selection to follow-up. Pretreatment “Pre-Radiomics”, intra-treatment “Intra-Radiomics” and post-treatment “Post-Radiomics” features could be used in the flow of radiotherapy treatment. Pre-Radiomics can be used in patient and treatment selection and radiotherapy planning. Intra-Radiomics has applications in treatment delivery, treatment verification and in conceptual adaptive radiotherapy (ART). Evaluation of toxicity, response to treatment can be quantified with the help of Post-Radiomics features. Unlike radiomics, deep learning is a form of AI that is based on a model similar to the structure of brain neurons, using several layers that allow the machine a continuous learning process. The radiomics potential and the perspectives for the clinical integration of the method in the diagnostic and clinical practice of HNC are also mentioned by Werth and collaborators (2-10).

Radiomics is already intensively evaluated in HNC, radiomics data being mixed with radiogenomic data. Radiogenomics represents the correlation of some features extracted from medical images with genomic data obtained from analysis of tissue. Although it is not the subject of this study, we will briefly mention some concepts of the use of radiomics and radiogenomics for the diagnosis of HNC: prediction of pathological staging, histological grade, nodal metastases and extranodal extension, perineural invasion, lymphovascular invasion and Human Papilloma Virus (HPV) status of HNC. Intensity, size, shape, texture and features corrected by applying filters are identified as the most frequently features correlated with the mentioned diagnostic data (11-12).

2. Radiomics in chemo-radiotherapy response prediction

Head and neck squamous cell carcinoma (HNSCC) represents the vast majority of HNC. Recurrence rates of over 50%, much higher in cases of locally advanced disease, make it necessary to identify innovative strategies for a stratified treatment approach. Currently, the standard treatment is concurrent chemoradiotherapy with Cisplatin weekly or every 3 weeks. Cases not eligible for platinum salts treatment could benefit from bio-radiotherapy with Cetuximab, an anti-epidermal growth factor receptor (EGFR) agent. Induction chemotherapy regimens followed by concurrent chemoradiotherapy are considered feasible in order to reduce the rate of distant recurrences and also to reduce the tumor volumes in bulky disease case. However, both the combination platinum – fluorouracil (PF) and taxanes – plantinum – fluorouracil (TPF) could be associated with an increased rate of toxic death (5.5% and 2.3% respectively). At the same time, for the subtype of oropharyngeal cancers associated with Human Papilloma Virus (HPV), treatment de-escalation strategies in order to reduce the rates of late toxic effects are currently being investigated (13,14).

Pre-treatment CT radiomics and whole transcriptome data were extracted, from a group of 206 HNSCC patients treated with chemo-radiotherapy. The aim of the study being to develop gene-based surrogate radiomic signatures in order to build a predictive model of loco-regional treatment response. The combined model including radiomics and transcriptomics demonstrated superior predictive capacity compared to the radiomic model, and the radiomic model was superior to the model based only on the genetic signature. Cone beam computer tomography (CBCT) imaging is routinely used in the modern concept of image guided radiotherapy (IGRT) to reduce ballistic errors in dose delivery. 40–80% reductions in margin from clinical target volume (CTV) to planning target volume (PTV) is feasible by using CBCT as routine in daily quality assurance (QA) of treatment. A reduction of the CTV-PTV margins from 5mm to 3mm with at least 1Gy limitation of the dose received by organs at risk (OARs) is also reported in cases of daily CBCT QA routine (15-18).

The pretreatment CT texture analysis in order to predict the failure of the chemoradiotherapy treatment in HNSCC was evaluated in a study that included 62 patients and the textural analysis was based on contrast enhanced CT images. 13 radiomic features were associated with local failure. The univariate analysis identified histogram, gray-level cooccurrence matrix, gray-level run-length, graylevel gradient matrix, and Laws features. The multivariate analysis identified 3 histogram features and 4 gray-level run-length features correlated with HNSCC response. Histogram statistics including maximum, mean, standard deviation, kurtosis and skewness were analyzed from pretreatment images in hypo-pharyngeal cancer and primary gross target volume (GTV-T) and nodal gross tumor volumes (GTV-N) were used as volumes of interest (VOI). In the group that included 20 supra-glottic and 5 pyriform sinus cases radiomic features pretreatment acquired could, predicted locoregional recurrence at 1 and 2 years. It should be mentioned that the subgroup analysis highlighted differences between radiomic features that predicted local recurrence in cases that received upfront radiotherapy (with or without concurrent chemotherapy) (19).

3. Radiomics and treatment related toxicities

Xerostomia and significantly reduction of the saliva quality is one of the main toxic effects of HNC radiotherapy (especially of the oral cavity and oropharynx). Although it is a multifactorial effect, the irradiation of the parotid glands, the submandibular glands, but also the minor salivary glands is involved in treatment related xerostomia. Chronic dysphagia with risk of aspiration pneumonia, trismus, mucositis, but also radio-necrosis of the temporal lobes (especially associated with radiotherapy of the skull base region) are also related with HNC treatment (20,21).

A study that aimed to evaluate radiomics as a predictor of xerostomia in HNC included 109 patients treated by helical tomo-therapy with radiation doses of 50-70Gy in 20-35 fractions. The parotid and submandibular glands were delineated as VOIs on the treatment planning CT. Moderate-to-severe xerostomia and sticky saliva were considered endpoints of the study 12 months post radiotherapy. Radiomic features Short Run Emphasis (SRE) and maximum CT intensity was evaluated as possible predictors of toxicities. Authors noted that these two radiomic features reported as predictive of xerostomia in other studies did not confirm any correlation in this case, mentioning the need for external validation of the radiomic models. 17 combined features (including 10 radiomics, 4 clinical and 3 DVH criteria) demonstrated the ability to predict saliva amount reduction. A systematic review that investigated MEDLINE/PubMed and EMBASE databases until June 2019 proposed by Carbonara et al. included studies of radiomics and AI in HNC (23) treatment related toxicity prediction. Radiomics quality score (RQS), a tool designed to assess radiomics methodology quality was also assessed. Among the 8 studies evaluated by 4 experts, only one radiomic analysis obtained an average RQS (considered ≤ 30%), and 3 studies obtained a score close to the average (≤ 25%). Radiomics analysis involved the parotid glands, cochlea, masticatory muscles and brain white matter as VOIs. The authors consider the results encouraging, needing additional validation in order to be able to be implemented as a tool for predicting the toxicities associated with irradiation (22-24).

4. Radiomics and ART

The new irradiation techniques deliver a homogeneous, tumoricidal dose to a welldefined region while minimizing the toxic effects in healthy tissues. The CT-simulator has become a standard in radiotherapy planning based on the definition of target volumes and radiosensitive structures. The possibility to fuse multiple structural or functional imaging methods with rigid or deformable algorithms opens up horizons both for the concept of “biological dose painting” and for the extensive use of radiomics in the planning stage of radiotherapy. Predicting the response to radiation treatment, both of the tumor and of the radiosensitive organs, is essential in the era of personalized medicine. The tumor control probability (TCP) and normal tissue complication probability (NTCP) radiobiological models are essential in optimizing treatment plans, but also in evaluating the possibility of HNC re-irradiation. Significant anatomical variations could have severe consequences on dosimetry and indirectly could be associated with underdosing or escalating the radiation doses in certain regions. For example, in HNC, the identification before treatment administration of cases that will have early parotid gland shrinkage could identify cases that require treatment re-planning. Radiomics and textural evaluation could have biomarker value for the necessity of adaptive radiotherapy (ART). Radiomic analysis of pre-treatment contrastenhanced magnetic resonance imaging (MRI) images could anticipate the consequences of anatomical and geometric variations in the planning of nasopharyngeal carcinoma by identifying the need for ART. The study that included 70 patients was based on 479 shape and texture radiomic features extracted from the GTV’s. Patients were randomly divided in two groups, one for model testing and one for validation. The Least Absolute Shrinkage and Selection Operator (LASSO) were used for the construction of the radiomic model. 13 out of 70 patients required ART and 6 radiomic features extracted from T1 and T2 MRI sequences were selected to create the optimal model. The authors mention the advantage of a model that includes both T1 and T2 sequences compared to models that use radiomic features extracted only from T1 or T2 MRI sequences (25-26).

Tumor hypoxia is one of the main causes of treatment failure. Positron emission tomography (PET) could identify tumor hypoxia regions using specific radiotracers based on 2-nitroimidazole structure. Dolezel and colleagues consider that an association between radiomics, genomic data and the concepts of dose painting by contour (DPBC) or by number (DPBN) using 18F-labeled fluoromisonidazole (18F-FMISO) will be the basis of ART in the future, being a cornerstone in of delivering precision radiotherapy. The pre-treatment evaluation of radiomic features extracted from CT imaging was evaluated on a group of 72 patients of which 36 required ART. Six features (4 semantic and 2 radiomic) generated the most accurate pre-treatment model for prediction of ART necessity in HNC. Ill-fitted thermoplastic masks (IfTMs) and the need for ART was predicted pre-treatment from CT-radiomics extracted from lymph nodes in nasopharyngeal cancer, starting from the idea that significant lymph node shrinkage is common during radiotherapy treatment. The results were

validated and confirmed in a multicenter study on 2 groups of 124 and 58 cases respectively (27-32). A brief presentation of several radiomic studies involving radiotherapy in head and neck cancers is summarized in Table 1.

Table 1. A brief presentation of several radiomic studies involving radiotherapy in head and neck cancers

5. Conclusions

In the future, in addition to the already intensively studied diagnostic applications, radiomics could be used for pre-therapeutic evaluation of the need for ART, predicting the response to radio-chemotherapy and anticipating treatment related toxicities. Clinical-radiomic models have superior predictive power and the delta variation of radiomic features could be a biomarker currently not enough investigated. Due to the characteristics of modern radiotherapy which includes as standard the IGRT concept using the CT simulator and the CBCT in order to ensure the accuracy of the patient’s positioning during the treatment, radiomics may soon be introduced in routine clinical practice and spearhead the HNC RGRT concept development.

Abbreviations:

AI – artificial intelligence

QA – quality assurance

HNC – head and neck cancer

ART – adaptive radiotherapy

IGRT – image guided radiotherapy

CT – computer tomography

CBCT – Cone Beam CT

RGRT – radiomic guided radiotherapy

HPV – Human Papilloma Virus

HNSCC – Head and neck squamous cell carcinoma

EGFR – epidermal growth factor receptor

PF – platinum – fluorouracil

TPF – taxanes – plantina – fluorouracil

HPV – Human Papilloma Virus

CTV – clinical target volume

PTV – planning target volume

OARs – organs at risk

GTV-T – primary gross target volume

GTV-N – nodal gross tumor volumes

VOI – volume of interest

SRE – Short Run Emphasis

RQS – Radiomics quality score

TCP – tumor control probability

NTCP – normal tissue complication probability

ART – adaptive radiotherapy

MRI – magnetic resonance imaging

LASSO – Least Absolute Shrinkage and Selection Operator

PET – Positron emission tomography

DPBC – dose painting by contour

DPBN – dose painting by number

18F-FMISO – 18F-labeled fluoromisonidazole

IfTMs – ill-fitted thermoplastic masks

Statements:

Author’s contributions: CCM, RII, and DTPI conceived of the discussed topic; CCM drafted the initial paper; RII and DTPI made the final revisions to the final paper.

Consent for publication: As the corresponding author, I confirm that the manuscript has been read by and approved for submission by all authors.

Funding: This study did not receive any specific grant from the funding agencies in the public, commercial, or not-for-profit sector.

Conflicts of Interest: The authors declare no conflict of interest.

References:

1. Siddique S, Chow JCL. Artificial intelligence in radiotherapy. Rep Pract Oncol Radiother. 2020 Jul-Aug;25(4):656-666. doi: 10.1016/j.rpor.2020.03.015.
2. Gillies RJ, Kinahan PE, Hricak H. Radiomics: Images Are More than Pictures, They Are Data. Radiology. 2016 Feb;278(2):563-77. doi: 10.1148/radiol.2015151169.
3. https://srrom.ro/layout/pdf/congres32/ProgramCongres2021.pdf
4. Zaharchuk G, Gong E, Wintermark M, Rubin D, Langlotz CP. Deep Learning in Neuroradiology. AJNR Am J Neuroradiol. 2018 Oct;39(10):1776-1784. doi: 10.3174/ajnr.A5543.
5. Kaka H, Zhang E, Khan N. Artificial Intelligence and Deep Learning in Neuroradiology: Exploring the New Frontier. Can Assoc Radiol J. 2021 Feb;72(1):35-44. doi: 10.1177/0846537120954293.
6. Werth K, Ledbetter L. Artificial Intelligence in Head and Neck Imaging: A Glimpse into the Future. Neuroimaging Clin N Am. 2020 Aug;30(3):359-368. doi: 10.1016/j.nic.2020.04.004.
7. Rizzo S, Botta F, Raimondi S, Origgi D, Fanciullo C, Morganti AG, Bellomi M. Radiomics: the facts and the challenges of image analysis. Eur Radiol Exp. 2018 Nov 14;2(1):36. doi: 10.1186/s41747-018-0068-z.
8. van Timmeren JE, Cester D, Tanadini-Lang S, Alkadhi H, Baessler B. Radiomics in medical imaging-“how-to” guide and critical reflection. Insights Imaging. 2020 Aug 12;11(1):91. doi: 10.1186/s13244-020-00887-2.
9. Duron L, Balvay D, Vande Perre S, Bouchouicha A, Savatovsky J, Sadik JC, Thomassin-Naggara I, Fournier L, Lecler A. Gray-level discretization impacts reproducible MRI radiomics texture features. PLoS One. 2019 Mar 7;14(3):e0213459. doi: 10.1371/journal.pone.0213459.
10. Zwanenburg A, Vallières M, Abdalah MA, Aerts HJWL, Andrearczyk V, Apte A, Ashrafinia S, Bakas S, Beukinga RJ, Boellaard R, Bogowicz M, Boldrini L, Buvat I, Cook GJR, Davatzikos C, Depeursinge A, Desseroit MC, Dinapoli N, Dinh CV, Echegaray S, El Naqa I, Fedorov AY, Gatta R, Gillies RJ, Goh V, Götz M, Guckenberger M, Ha SM, Hatt M, Isensee F, Lambin P, Leger S, Leijenaar RTH, Lenkowicz J, Lippert F, Losnegård A, Maier-Hein KH, Morin O, Müller H, Napel S, Nioche C, Orlhac F, Pati S, Pfaehler EAG, Rahmim A, Rao AUK, Scherer J, Siddique MM, Sijtsema NM, Socarras Fernandez J, Spezi E, Steenbakkers RJHM, Tanadini-Lang S, Thorwarth D, Troost EGC, Upadhaya T, Valentini V, van Dijk LV, van Griethuysen J, van Velden FHP, Whybra P, Richter C, Löck S. The Image Biomarker Standardization Initiative: Standardized Quantitative Radiomics for High-Throughput Image-based Phenotyping. Radiology. 2020 May;295(2):328-338. doi: 10.1148/radiol.2020191145.
11. Mukherjee P, Cintra M, Huang C, Zhou M, Zhu S, Colevas AD, Fischbein N, Gevaert O. CT-based Radiomic Signatures for Predicting Histopathologic Features in Head and Neck Squamous Cell Carcinoma. Radiol Imaging Cancer. 2020 May 15;2(3):e190039. doi: 10.1148/rycan.2020190039.
12. Gul M, Bonjoc KC, Gorlin D, Wong CW, Salem A, La V, Filippov A, Chaudhry A, Imam MH, Chaudhry AA. Diagnostic Utility of Radiomics in Thyroid and Head and Neck Cancers. Front Oncol. 2021 Jul 7;11:639326. doi: 10.3389/fonc.2021.639326.
13. Vermorken JB, Remenar E, van Herpen C, Gorlia T, Mesia R, Degardin M, Stewart JS, Jelic S, Betka J, Preiss JH, van den Weyngaert D, Awada A, Cupissol D, Kienzer HR, Rey A, Desaunois I, Bernier J, Lefebvre JL; EORTC 24971/TAX 323 Study Group. Cisplatin, fluorouracil, and docetaxel in unresectable head and neck cancer. N Engl J Med. 2007 Oct 25;357(17):1695-704. doi: 10.1056/NEJMoa071028.
14. Economopoulou P, Kotsantis I, Psyrri A. De-Escalating Strategies in HPV-Associated Head and Neck Squamous Cell Carcinoma. Viruses. 2021 Sep 8;13(9):1787. doi: 10.3390/v13091787.
15. Rabasco Meneghetti A, Zwanenburg A, Linge A, Lohaus F, Grosser M, Baretton GB, Kalinauskaite G, Tinhofer I, Guberina M, Stuschke M, Balermpas P, von der Grün J, Ganswindt U, Belka C, Peeken JC, Combs SE, Böke S, Zips D, Troost EGC, Krause M, Baumann M, Löck S. Integrated radiogenomics analyses allow for subtype classification and improved outcome prognosis of patients with locally advanced HNSCC. Sci Rep. 2022 Oct 6;12(1):16755. doi: 10.1038/s41598-022-21159-7.
16. Kearney M, Coffey M, Leong A. A review of Image Guided Radiation Therapy in head and neck cancer from 2009- 201 – Best Practice Recommendations for RTTs in the Clinic. Tech Innov Patient Support Radiat Oncol. 2020 Jun 12;14:43-50. doi: 10.1016/j.tipsro.2020.02.002.
17. van Kranen S, Hamming-Vrieze O, Wolf A, Damen E, van Herk M, Sonke JJ. Head and Neck Margin Reduction With Adaptive Radiation Therapy: Robustness of Treatment Plans Against Anatomy Changes. Int J Radiat Oncol Biol Phys. 2016 Nov 1;96(3):653-60. doi: 10.1016/j.ijrobp.2016.07.011.
18. 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 Jan;61(1):73-80. doi: 10.1080/0284186X.2021.1983207.
19. Bahig H, Lapointe A, Bedwani S, de Guise J, Lambert L, Filion E, Roberge D, Létourneau-Guillon L, Blais D, Ng SP, Nguyen-Tan PF. Dual-energy computed tomography for prediction of loco-regional recurrence after radiotherapy in larynx and hypopharynx squamous cell carcinoma. Eur J Radiol. 2019 Jan;110:1-6. doi: 10.1016/j.ejrad.2018.11.005.
20. Kaae JK, Stenfeldt L, Eriksen JG. Xerostomia after Radiotherapy for Oral and Oropharyngeal Cancer: Increasing Salivary Flow with Tasteless Sugar-free Chewing Gum. Front Oncol. 2016 May 3;6:111. doi: 10.3389/fonc.2016.00111.
21. Mireștean, C., Buzea, C., Butuc, I., Zara, A., & Iancu, D. (2017). Comparative evaluation of the doses received by the parotid glands as predictors of xerostomia be 3D-CRT, IMRT and VMAT irradiation techniques in local advanced nasopharynx cancer. Archive of Clinical Cases, 4(3), Arch Clin Cases 2017; 4(3):146-153. doi:http://dx.doi.org/10.22551/2017.16.0403.10107
22. Azadikhah A, Varghese BA, Lei X, Martin-King C, Cen SY, Duddalwar VA. Radiomics quality score in renal masses: a systematic assessment on current literature. Br J Radiol. 2022 Sep 1;95(1137):20211211. doi: 10.1259/bjr.20211211.
23. Berger T, Noble DJ, Yang Z, Shelley LEA, McMullan T, Bates A, Thomas S, Carruthers LJ, Beckett G, Duffton A, Paterson C, Jena R, McLaren DB, Burnet NG, Nailon WH. Assessing the generalisability of radiomics features previously identified as predictive of radiation-induced sticky saliva and xerostomia. Phys Imaging Radiat Oncol. 2022 Dec 16;25:100404. doi: 10.1016/j.phro.2022.12.001.
24. Carbonara R, Bonomo P, Di Rito A, Didonna V, Gregucci F, Ciliberti MP, Surgo A, Bonaparte I, Fiorentino A, Sardaro A. Investigation of Radiation-Induced Toxicity in Head and Neck Cancer Patients through Radiomics and Machine Learning: A Systematic Review. J Oncol. 2021 Jun 9;2021:5566508. doi: 10.1155/2021/5566508.
25. Iancu DT, Iancu RI. (Use of simulator in treatment planning and determination and definition of treatment volume). Revista Medico-chirurgicala a Societatii de Medici si Naturalisti din Iasi. 2004 Jul-Sep;108(3):580-583.
26. Yu TT, Lam SK, To LH, Tse KY, Cheng NY, Fan YN, Lo CL, Or KW, Chan ML, Hui KC, Chan FC, Hui WM, Ngai LK, Lee FK, Au KH, Yip CW, Zhang Y, Cai J. Pretreatment Prediction of Adaptive Radiation Therapy Eligibility Using MRI-Based Radiomics for Advanced Nasopharyngeal Carcinoma Patients. Front Oncol. 2019 Oct 16;9:1050. doi: 10.3389/fonc.2019.01050.
27. Rasey JS, Grunbaum Z, Magee S, Nelson NJ, Olive PL, Durand RE, Krohn KA. Characterization of radiolabeled fluoromisonidazole as a probe for hypoxic cells. Radiat Res. 1987 Aug;111(2):292-304.
28. Iancu DT, Iancu RI. (Use of simulator in treatment planning and determination and definition of treatment volume). Revista Medico-chirurgicala a Societatii de Medici si Naturalisti din Iasi. 2004 Jul-Sep;108(3):580-583.
29. Dolezel M, Slavik M, Blazek T, Kazda T, Koranda P, Veverkova L, Burkon P, Cvek J. FMISO-Based Adaptive Radiotherapy in Head and Neck Cancer. J Pers Med. 2022 Jul 29;12(8):1245. doi: 10.3390/jpm12081245.
30. Brown E, Owen R, Harden F, Mengersen K, Oestreich K, Houghton W, Poulsen M, Harris S, Lin C, Porceddu S. Predicting the need for adaptive radiotherapy in head and neck cancer. Radiother Oncol. 2015 Jul;116(1):57-63. doi: 10.1016/j.radonc.2015.06.025.
31. Lam SK, Zhang J, Zhang YP, Li B, Ni RY, Zhou T, Peng T, Cheung AL, Chau TC, Lee FK, Yip CW, Au KH, Lee VH, Chang AT, Chan LW, Cai J. A Multi-Center Study of CT-Based Neck Nodal Radiomics for Predicting an Adaptive Radiotherapy Trigger of Ill-Fitted Thermoplastic Masks in Patients with Nasopharyngeal Carcinoma. Life (Basel). 2022 Feb 6;12(2):241. doi: 10.3390/life12020241.
32. C. C. Mireştean, A. D. Zară, R. I. Iancu and D. P. T. Iancu, “CT Re-Simulation Based Delta-Radiomics New Perspectives Beyond Adaptive Radiotherapy (ART),” 2021 International Conference on e-Health and Bioengineering (EHB), Iasi, Romania, 2021, pp. 1-4, doi: 10.1109/EHB52898.2021.965764