Proton beam pauses mid-heartbeat to spare healthy tissue in cardiac ablation

Every heartbeat is a moving target. When radiation oncologists attempt to destroy the rogue electrical circuits that trigger ventricular tachycardia, they must fire a massive dose of radiation at a piece of heart muscle that contracts, relaxes, rises and falls with every breath and every beat. Now a team at Mass General Brigham and Harvard Medical School, working with collaborators at Yale University, EBAMed in Geneva and several industry partners, has validated a complete end-to-end workflow for delivering proton therapy to the heart with the beam switched off during most of the cardiac cycle. The study, published in Frontiers in Oncology, treated six healthy Yucatan swine and demonstrated that gating the beam to the brief diastolic window of the heartbeat shrank the radiation target’s motion-related expansion by an average of 45 percent, substantially reducing the amount of healthy myocardium caught in the firing line.

Stereotactic arrhythmia radiotherapy, or STAR, is emerging as a non-invasive alternative to catheter ablation for ventricular tachycardia. Early clinical evidence has shown the approach is safe and can reduce a patient’s VT burden, and most patients to date have been treated with photon beams. Protons, however, offer a potential safety advantage: a proton beam deposits its dose at depth and then stops almost abruptly, thanks to its distal fall-off and high dose conformality, which can increase the sparing of healthy tissues in proximity to the target. But cardiac motion is complex. It has multiple components — heartbeat-related contraction, respiration-related displacement and slower tissue drift — each with patient-specific amplitudes that can reach several millimeters. Until now, no clinically available system could gate an external beam to real-time cardiorespiratory motion information.

The central technical achievement of the new study lies in the proton pencil beam scanning controller. Pencil beam scanning magnetically steers a narrow proton beam to paint the target layer by layer, spot by spot, with high precision — a fundamentally different approach from double-scattering, which physically spreads a wider beam across the target area. Respiratory gating had previously been implemented at Mass General Hospital in double-scattering mode, but no gating capability existed for pencil beam scanning at the institution. The problem is one of timescales: individual dose spots in a clinically relevant 25 Gy plan can take more than 300 milliseconds to deliver, while the diastolic gating window — the quiescent phase between 70 and 100 percent of the R-R interval — is shorter than that delivery time. A controller that only observes the gate signal at spot boundaries would spill dose outside the window. The upgraded system, built on a Pyramid Technical Consultants scanning controller that manages the beam independently of the upstream IBA cyclotron, can interrupt the beam and resume delivery mid-stream within an individual dose spot, splitting it across as many gate-off intervals as needed.

Latency measurements confirmed the system’s speed. Using a fast PC-based oscilloscope with 100 MHz analog bandwidth and 1 gigasample-per-second sampling, the team overlaid the CardioKit gate signal with the cyclotron integrated current monitor ion chamber. At maximum beam current, the ramp-up and ramp-down latency between a gate command and the rise or cessation of beam current was less than 550 microseconds, with full beam current ramp-up taking approximately 2 milliseconds and ramp-down approximately 1 millisecond. Both values are negligible relative to the roughly 50-millisecond per-spot delivery time. In functional testing on a dynamic thorax phantom, asynchronous motion between the detector and the beam spot during a uniform 40 by 50 square-millimeter field scan badly degraded the planned dose uniformity — a 1-second periodic, 1-centimeter peak-to-peak motion degraded uniformity by approximately 21.6 percent. Gated scanning with an approximate 30 percent duty cycle largely restored uniformity to 6.5 percent, demonstrating that the interplay effects between moving anatomy and a scanning beam can be effectively mitigated.

The second pillar of the workflow is CardioKit, an investigational ultrasound-based image-guidance system developed specifically for STAR by EBAMed. The device combines hands-free transthoracic echocardiographic imaging with electrocardiographic data streams, providing real-time soft-tissue images of the left ventricle before and during treatment. Optical tracking of the ultrasound probe on the patient’s thorax determines the target’s three-dimensional motion in the same coordinate system as the simulation imaging and treatment delivery systems — a crucial calibration step performed with a phantom containing internal radio-opaque ball bearings and external infrared-reflecting spheres. Unlike external surrogates such as the Real-time Position Management marker block, ultrasound visualizes the internal anatomy directly, in a non-ionizing way. Each ultrasound image is binned by cardiac phase and registered to the baseline image of the corresponding phase, computing respiratory displacement in three spatial coordinates without including heartbeat contraction itself.

CardioKit’s accuracy was benchmarked against respiratory 4D-CT, the state-of-the-art reference. For two cases, the Pearson correlation between the ultrasound-derived respiratory motion traces and the 4D-CT traces was 0.99 and 0.97, with differences in total breathing motion amplitude of roughly 1.1 to 1.4 millimeters. The system also proved remarkably fast at managing heartbeat irregularities. Simulator-based tests covering atrial fibrillation, premature atrial contractions, frequent multifocal premature ventricular contractions, bigeminy, trigeminy, pairs and runs of PVCs, and ventricular tachycardia — across all ECG lead configurations — showed that CardioKit detected every simulated arrhythmic event as an abnormally short heartbeat and interrupted the gating signal in under 5 milliseconds. That response time is less than 1 percent of the cardiac phase at the study’s average treatment heart rate of 102 beats per minute, and finer than the 5 percent R-R phase resolution of the cardiac 4D-CT used to model cardiac motion at simulation.

The treatment planning comparison between the two gating strategies revealed the dosimetric payoff. Cardiac target volumes, defined at the quiescent 80 percent R-R phase by consensus of a radiation oncologist and cardiologist, encompassed the left ventricle free wall and papillary muscle. For dual-gated subjects, the internal target volume expansion accounted only for motion within the diastolic window, yielding expansions of 16 to 23 percent. For single-gated subjects, whose beam delivery spanned the full cardiac cycle, expansions reached 48 to 73 percent — a mean of 19 percent versus 64 percent, the 45 percent reduction headline. After adding a standard 5-millimeter isotropic planning margin for setup uncertainty and residual respiratory motion, the final planning target volume was geometrically magnified to roughly 4.1 to 5.8 times the initial target volume, underscoring how disproportionately severe the volumetric penalty becomes for smaller targets when cardiac gating is absent. All plans delivered 35 GyRBE in a single fraction using two pencil beam scanning fields with 12-centimeter apertures and energies from 100 to 150 MeV, while enforcing a maximum dose limit of 15 GyRBE to the left anterior descending coronary artery.

Gated delivery came at the cost of beam-on time, and the delivery logs tell the story of the mid-spot interrupt capability working as designed. For dual-gated subjects, the delay factor — the ratio of gated to non-gated beam-on time — averaged 6.6, with total gated beam-on times of 1431, 677 and 1037 seconds across two beams per animal. Single-gated subjects averaged a delay factor of 2.7. In the dual-gated group, 56.3 percent of spots experienced at least one gating interruption, compared with 44.0 percent in the single-gated group, and among gated spots roughly 42 percent needed multiple interruptions before accumulating their prescribed monitor units. Average gated spot delivery time ballooned to 778 milliseconds in the dual-gated group versus 294 milliseconds in the single-gated group, with individual maxima of 10055 and 3807 milliseconds. Critically, no treatment aborts due to gating timeout were recorded in either cohort, and no CardioKit device failure from radiation exposure occurred throughout the study.

The authors are careful about limitations. The study used healthy swine rather than arrhythmia patients, and the animal model has shortcomings: the swine thorax is more curved than the relatively flatter human thorax, the heart’s respiratory motion is much smaller, and the heartbeat is significantly faster than in humans — all of which made ultrasound probe placement harder and reduced the gating duty cycle compared with what patients would experience. The animals were also sedated and mechanically ventilated, which enabled comparison of non-synchronous 4D-CT, RPM and CardioKit motion curves but influenced duty cycle results. Nevertheless, the swine model is standard practice in electrophysiology and STAR research, and the workflow — from stoichiometric CT calibration and ECG-gated 4D-CT simulation, through contouring, planning and gating window definition, to gated irradiation — comprises steps that modern radiation therapy departments can accomplish. The team notes that robustness optimization and interplay calculations would likely replace the uniform planning margin in a clinical context, and that more advanced patient positioning imaging could shrink the setup margin below 5 millimeters. As an end-to-end validation of cardiorespiratory-gated proton pencil beam scanning for cardiac ablation, the study lays the groundwork for bringing this motion-aware, tissue-sparing approach to human patients with ventricular tachycardia.

Subject of Research: Cardiorespiratory-gated proton pencil beam scanning for cardiac ablation

Article Title: Cardio-respiratory-gated proton pencil beam scanning for cardiac ablation: workflow and end-to-end validation

Article References: Depauw, N., Bijari, F., Cascio, E., Driscoll, E., Garonna, A., Kang, J., Deng, H., Kim, H., Macdonald-Soccorso, M., Nett, W., Normandin, M., Pellissier-Tanon, M., Woodcock, R., Yaghoubian, K., El Fakhri, G., Levine, R., Lehmann, H. I., Chen, Y.-L., & Jee, K.-W. (2026). Cardio-respiratory-gated proton pencil beam scanning for cardiac ablation: workflow and end-to-end validation. Frontiers in Oncology, 16, Article 1923998. https://doi.org/10.3389/fonc.2026.1923998

Image Credits: AI Generated

DOI: 10.3389/fonc.2026.1923998

Keywords: proton therapy, cardiac ablation, ventricular tachycardia, stereotactic arrhythmia radiotherapy, motion management, pencil beam scanning, ultrasound image guidance, gating, CardioKit, swine study, 4D-CT, interplay effects