ESTRO 2021 Abstract Book

S458

ESTRO 2021

1 University Medical Center Utrecht, Department of Radiotherapy, Utrecht, The Netherlands

Abstract Text Ventricular tachycardia (VT) is one of the main contributing factors to sudden cardiac death (SCD). VT describes a sudden acceleration in the heart rhythm up to a level where the normal blood pumping function of the ventricles cannot be maintained. It is indicative of a disruption or disturbance of the normal cardio-electrical circuit induced by myocardial scar tissue. In Europe, the annual incidence of SCD is approximately 50-100 per 100,000 inhabitants [Fishman 2010]. This corresponds to approximately 500,000 cases of SCD per year. The current standard-of-care is to use cardioprotective and anti-arrhythmic drugs, anti- tachycardia pacing and/or shocks delivered by an implantable cardioverter-defibrillator (ICD), and invasive catheter ablation for drug-refractory VT. Unfortunately, VT recurs in 30-50% of all patients within one year of catheter ablation. Very recently, advances in both cardiology and radiotherapy have given rise to a new non-invasive salvage treatment for VT: STereotactic Arrhythmia Radioablation (STAR). STAR has been named in analogy to Stereotactic Ablative Body Radiotherapy (SABR). Typically, a single fraction dose of 25 Gy is applied to ablate the ventricular target [Cuculich 2017]. It is estimated that <200 patients have been treated so far worldwide, mostly under compassionate care protocols, or limited phase I/II trials. Remarkably, despite the ad-hoc nature of cardiac radioablation, STAR has reportedly reduced the VT burden by >90% with a favourable toxicity profile [Robinson 2019]. Within Europe, we have initiated the H2020-funded STOPSTORM consortium (#945119), in an effort to consolidate all European cardiac radioablation initiatives. By pooling data from more than 30 participating centres, we hope to comprehensively demonstrate the safety and efficacy of STAR. Cleary, the effectiveness of STAR hinges on the accuracy of targeting, and the precision of the irradiation. Ideally, treatment would eradicate the arrhythmia focus while also avoiding radiation-induced toxicities in healthy tissue. This is especially difficult in the heart, which is subject to complex cardiorespiratory motion, while itself being considered a very important organ-at-risk. In this presentation, I will address the three key factors ( 3Ts ) that could impede the widespread adoption of STAR. Targeting : Identifying the relevant electrophysiological (EP) substrate for STAR is the first key challenge. Treatment : Ablating the EP target, subject to complexcardiorespiratory motion, while minimizing radiation-induced toxicities is the second key challenge. Toxicity : Keeping track of acute and longer-term radiation-induced side effects of STAR is the third key challenge. A special focus of this presentation will be on STAR using hybrid MRI/radiotherapy (MR-linac) devices to maximize treatment precision by tracking patients’ every heartbeat and breath during irradiation. References: [Fishman 2010]: DOI: 10.1161/CIRCULATIONAHA.110.976092 [Cuculich 2017]: DOI: 10.1056/NEJMoa1613773 [Robinson 2019]: DOI: 10.1161/CIRCULATIONAHA.118.038261 Machine learning and analysis of data produced by radiation oncology is a relatively new and fascinating field. However, models based on databases needs to be understood by the end users, mainly clinicians, to encourage them to implement these tools in daily clinical practice. Despite the vast amount of publication about using ML in analyzing multisource data, there is still some obstacles in both performance of such predictive models and their seamless inclusion in radiotherapy workflow. Although ML has great potential to improve the lives of patients and clinicians, developing new ML tools is a challenging process. The most important first step is defining the clinical question or problem to be tackled, e.g. decision support, prognosis and prediction or generating new hypotheses for further research. At this stage the involvement of end users, for example clinicians, is crucial to identify and frame the goals of the project. Apart from clinicians, the field of clinical ML involves researchers from different disciplines: computer scientists, medical physicists, and biostatisticians. Communication and cooperation between the algorithm developers and domain experts is very important to fully understand both the context and limitations of the data. The potential of multisource databases seems to be endless, due to possibility of analysing relationships between many features and use powerful ML algorithms to find novel correlations and interactions. However, to build a credible ML tool it's often worth to start with the simplest possible approach and gradually build up the complexity while benchmarking the performance against a well-defined set of baselines. This makes it possible to demonstrate the added value of a more complex ML approach, as simple features and known predictive factors like for example tumour volume can often perform surprisingly well. Additionally, best practices and pitfalls in database preparation and model validation will be discussed, as well as issues of data sharing and ethical considerations including the responsibility for decisions made based on algorithmic suggestions. SP-0592 Learning machine learning J. Kazmierska 1 1 Greater Poland Cancer Center, Radiotherapy Department II, Poanan, Poland Abstract Text

SP-0593 How to get a new topic published? (from an editor about how to publish new types of research, level of detail for still-unfamiliar audience.) TBC

Abstract not available