Patient-specific quality assurance, or PSQA, checks whether the dose delivered during radiotherapy agrees with the approved treatment plan. EPID-based methods are attractive because the imaging panel is integrated into the accelerator and can acquire a two-dimensional record of the beam actually emerging from the treatment head. Such images can be compared with calculated portal-dose images and evaluated with gamma analysis, producing both a summary pass rate and a spatial display of discrepancies. The drawback is that conventional EPID PSQA is generally performed once, before treatment begins, so its result describes the machine’s condition on that particular day rather than its condition for every subsequent fraction.

The study addressed that limitation by training a Pix2Pix conditional generative adversarial network to translate an Anisotropic Analytical Algorithm portal-dose calculation into a predicted EPID image. The authors tested two output designs: one that predicted the full EPID intensity and another that predicted the dose-difference map between calculated and measured images. They also compared three input and training configurations: a baseline using only the calculated image, a model conditioned on daily Machine Performance Check data, and a third model that added synthetic multi-leaf collimator deviations to its training set. These axes produced six model variants, all built with the same U-Net-style generator and trained for 2,000 epochs.

The machine-state input consisted of 235 features selected from 253 values recorded by the Halcyon system’s automated daily quality-assurance procedure. Most were related to the positions and backlash of the 57 leaves in each bank of the dual-layer multi-leaf collimator. Seven additional values described beam output, uniformity, field edges, collimator rotation and EPID calibration. Each deviation was normalized by its own quality-control tolerance, allowing features expressed in millimetres, percentages and degrees to enter the network on comparable scales. The vector was projected into a 4096-dimensional embedding, reshaped into a feature map and stacked with the calculated portal-dose image before processing by the U-Net generator.

The dataset comprised 135 normally delivered arc fields from 50 patients treated between April 2023 and October 2024. The patients included eight head-and-neck cases, 34 pelvic cases, of which 26 were prostate treatments, six thorax cases and two extremity cases. All plans used two or three arcs and the 6 MV flattening-filter-free beam. Patient-level separation kept all arcs from a patient in the same subset: 107 fields from 39 patients were used for training and 28 fields from 11 patients formed the common test set. Images were resampled to a uniform 256-by-256 matrix at 1 mm spacing, with no rotation, translation or cropping augmentation.

Because clinically delivered plans rarely contain large collimator errors, the researchers also created 36 deliberately perturbed fields. They selected six base arcs and applied six preset plus-or-minus 1 mm positioning patterns, including asymmetric single-bank shifts, symmetric aperture widening or narrowing, and localized changes affecting selected leaves. The modified plans were physically re-delivered and measured on the accelerator. Corresponding leaf positions in the machine-performance vectors were adjusted by the same offsets, producing paired examples in which the recorded machine state and measured image changed consistently. Twenty-six perturbed fields were added to training, while 10 were combined with their six unperturbed source arcs in a separate plan-paired evaluation set.

On the common 28-field test set, the full difference-target model achieved the best overall agreement with measured gamma passing rates. At 3%/2 mm, its root-mean-square error was 0.48 percentage points and its correlation was 0.944, compared with errors of 0.69 and 0.55 percentage points for the baseline difference-target and EPID-target models. At the stricter 2%/2 mm criterion, the full difference-target model recorded an error of 1.27 percentage points and a correlation of 0.936, while the corresponding baseline errors were 2.01 and 1.68 percentage points. However, bootstrap confidence intervals for the differences between the full and baseline difference-target models included zero, and an exact paired Wilcoxon test found no significant difference in per-field absolute errors on normal plans. The authors therefore characterized the numerical advantage on unperturbed cases as part of the sampling variability rather than confirmed superiority.

The clearest effect appeared in the deliberately perturbed fields. At 3%/2 mm, mean absolute gamma-passing-rate error in the difference-target chain fell from 3.67 percentage points in the baseline model to 2.94 points after adding machine-performance conditioning, then to 2.12 points after also adding deviation-aware augmentation. The full 42% reduction was statistically significant for the paired perturbed samples, with p equal to 0.020. The corresponding reduction in the EPID-target chain was smaller, from 3.45 to 2.71 percentage points, and adding machine-state information alone reversed the direction of that chain’s prediction error. Feature-masking experiments showed why: replacing the 228 MLC-related values with their nominal state increased the full model’s signed error on perturbed plans from positive 1.83 to positive 3.40 percentage points, while masking non-MLC groups had little effect.

Pixel-level detection remained substantially weaker than plan-level prediction. Across all model variants, sensitivity to gamma failures ranged from 19.65% to 32.18%, depending on the criterion and configuration, meaning that most failing pixels were not identified. Specificity and accuracy were high, reaching as much as 99.88% and 99.67%, respectively, but the researchers cautioned that a predicted gamma passing rate cannot replace physical measurement or reliably localize every failure region. The study was also restricted to one institution and one Halcyon v3.0 platform, its perturbation set contained only 1 mm collimator-position offsets, and the plan-paired test set was small. The results therefore establish retrospective agreement and perturbation sensitivity, not clinical readiness, and the authors call for prospective testing of error detection, false-negative risk and stability across treatment fractions.

Subject of Research: Deep learning for machine-state-aware prediction of radiotherapy delivery quality using EPID-based patient-specific quality assurance

Article Title: A deep learning approach for EPID-based patient-specific quality assurance incorporating daily machine-state variations

Article References: Pak, K., Kim, S.-W., Park, K., Cho, B., Kim, Y. J., Song, S. Y., Jeong, C., & Kwak, J. (2026). A deep learning approach for EPID-based patient-specific quality assurance incorporating daily machine-state variations. Frontiers in Oncology, 16, Article 1911466. https://doi.org/10.3389/fonc.2026.1911466

Image Credits: AI Generated

DOI: 10.3389/fonc.2026.1911466

Keywords: radiotherapy, patient-specific quality assurance, electronic portal imaging, deep learning, gamma passing rate, linear accelerator, multi-leaf collimator, generative adversarial network, machine performance check, deep, learning, approach

The population studied represents the hardest edge of a condition that affects millions. Treatment-resistant depression is generally defined as depression that fails to improve adequately after at least two antidepressant treatments, and it is estimated to affect roughly three in ten people with the illness. For these patients, the remaining options are limited and burdensome: additional medication trials, ketamine-based therapies, electroconvulsive therapy, or invasive ablative procedures such as radiofrequency ablation. Each carries meaningful risks of side effects, and many demand repeated sessions or actual surgery. It is against this backdrop that the Vanderbilt team, working with colleagues at the University of South Florida and Tampa General Hospital, explored whether the precision of modern radiation therapy could offer something new.

Stereotactic radiosurgical capsulotomy works by targeting a small region of the anterior limb of the internal capsule, a brain pathway long implicated in mood regulation and obsessive-compulsive symptoms. Capsulotomy itself has a decades-long history in the treatment of severe psychiatric illness, but the radiosurgical version has been studied almost exclusively in severe, treatment-resistant obsessive-compulsive disorder rather than depression. The procedure requires no incision and no implanted device. Instead, patients are fitted with a mask that immobilizes the head, and a linear accelerator delivers cross-fired radiation beams that converge on a target only millimeters wide. The goal, as lead author Dr. Mohamed Khattab, a radiation oncologist at the University of South Florida and Tampa General Hospital, explained, is to interrupt a circuit contributing to severe, persistent symptoms while sparing the surrounding brain tissue.

The new study builds on a small 2022 pilot in which three patients underwent frameless stereotactic radiosurgical capsulotomy for refractory depression. In that initial report, all three experienced reductions or complete resolution of suicidal thoughts, with no observed cognitive decline or neurologic side effects. Encouraged by those results, the team expanded the program. Between 2020 and 2025, twenty patients with treatment-resistant depression, ten of whom also carried a diagnosis of comorbid obsessive-compulsive disorder, were treated at Vanderbilt for medical necessity after review by a multidisciplinary team spanning psychiatry, neurosurgery and radiation oncology. The severity of their histories is difficult to overstate: every patient had previously received cognitive therapy and multiple medications, and many had failed to improve with ketamine, transcranial magnetic stimulation, electroconvulsive therapy, or some combination of these.

The results, measured with standardized instruments and quarterly evaluations over a median follow-up of 38.6 months, were striking. Among the 18 patients evaluated for depression symptoms, the average score on the Beck’s Depression Inventory, a widely used self-report measure, fell by 53 percent from before radiosurgery to the point of greatest improvement. Eleven patients, or 61 percent, saw their scores drop by at least half, and 16 patients, or 89 percent, showed some numerical improvement. Crucially, the benefits did not evaporate with time. At the most recent follow-up, the average depression score remained 47 percent below the pretreatment baseline, half of the evaluated patients still met the threshold for a treatment response, and 89 percent remained improved compared with where they started.

Perhaps the most dramatic number concerned suicidal ideation. Before treatment, 14 of the 20 patients, or 70 percent, reported active thoughts of suicide. After radiosurgery, exactly one patient, or 5 percent, did. For a population in which many individuals were living with persistent, treatment-refractory suicidal thinking, that shift represents a potentially life-altering outcome, and it is the aspect of the study that outside experts have flagged as most compelling. Dr. Khattab noted that these were patients whose depression had not improved after several other treatments and that many were living with active thoughts of suicide, yet after treatment the majority experienced meaningful reductions in both depressive symptoms and suicidal thoughts.

Equally important for a procedure that targets the brain is what did not happen. Researchers found no overall evidence of cognitive decline. Scores on the Montreal Cognitive Assessment were stable or numerically higher after treatment, with all 20 patients showing stable or improved scores at their point of greatest improvement and 19 of 20 doing so at last follow-up. The team also observed no neurologic deficits, no apathy and no personality changes, outcomes that address a common and understandable fear about any intervention involving brain tissue. Three patients did experience moderate treatment-related side effects consisting of headache or dizziness, and two of those also showed larger-than-intended tissue changes on imaging, but all recovered fully with steroid treatment.

Dr. Khattab and his colleagues are careful to frame the findings within their limits. The study was small and lacked a comparison group, which means it cannot establish that radiosurgery itself caused the improvements; spontaneous remission, placebo effects, the intense clinical attention participants received, or regression to the mean in patients selected at their lowest point could all contribute. Larger, controlled studies will be needed to confirm the results and to clarify which patients are most likely to benefit and what risks may emerge over longer horizons. Caution is especially warranted, the researchers argue, given the troubled history of ablative psychosurgery. Modern stereotactic techniques are vastly more precise than the procedures of the mid-twentieth century, but radiosurgical capsulotomy still demands rigorous patient selection, thorough informed consent and specialized multidisciplinary expertise. Although the linear accelerator technology used to deliver the treatment is widely available in radiation oncology departments, Dr. Khattab emphasized that the procedure itself should remain confined to experienced teams.

That sentiment is echoed within the radiation oncology community. Dr. Markus Bredel, co-chair of ASTRO’s Functional Radiation Medicine Task Force and chair of radiation oncology at the University of Miami Sylvester Comprehensive Cancer Center, described the study as important because it evaluates a highly targeted, incisionless approach in patients who had already tried multiple therapies while also tracking cognition and safety over time. The reduction in suicidal ideation, he observed, is striking, but the findings should be viewed as early evidence from a specialized, multidisciplinary program, and larger studies will be needed before the approach can be understood as part of routine care.

Even with those caveats, the study signals a notable convergence between radiation medicine and psychiatry. It suggests that the same physics that allows clinicians to ablate a tumor with submillimeter accuracy can be redirected toward dysfunctional neural circuits, offering an incisionless alternative for patients who have exhausted pharmacologic, stimulation-based and psychotherapeutic options. For the field, the immediate task is rigorous replication: randomized or matched controlled trials, longer follow-up, standardized target definition and clearer criteria for candidate selection. For patients who have lived for years with severe depression despite trying nearly every available treatment, as Dr. Khattab put it, the possibility of another option is meaningful, and these results give researchers a concrete reason to study the approach more rigorously so that those who may benefit can receive it as safely as possible.

Subject of Research: Stereotactic radiosurgical capsulotomy for treatment-resistant major depression

Article Title: Incisionless radiosurgery shows promise for severe, treatment-resistant depression

Article References: Incisionless radiosurgery shows promise for severe, treatment-resistant depression. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: treatment-resistant depression, stereotactic radiosurgery, capsulotomy, suicidal ideation, radiation oncology, neuromodulation, Beck Depression Inventory, cognitive outcomes, ASTRO, psychiatry, linear accelerator, clinical study

The research team, led by Jesse Lohela and colleagues at the University of Oulu, compared two TrueBeam machines fitted with HyperSight against two standard TrueBeam systems, using the widely recognized Catphan 504 imaging phantom. HyperSight brings a substantial hardware refresh to the imaging chain: a larger 43 by 43 centimeter detector panel replacing the standard 40 by 30 centimeter unit, a cesium iodide scintillator instead of gadolinium oxysulfide, a stronger anti-scatter grid with a ratio of 15:1 rather than 10:1, and a faster gantry rotation speed of 1.5 revolutions per minute compared with 1.0. Together these changes extend the usable field of view to as much as 70 centimeters, up from 46.5 centimeters, and enable updated iterative reconstruction algorithms with Acuros-based scatter correction and an optional metal artifact reduction mode.

The physics behind why CBCT image quality lags behind diagnostic CT is central to the study. The wide cone-shaped X-ray beam used in linac-mounted imaging generates far more scatter radiation than the narrow fan beam of a CT scanner, degrading contrast and distorting Hounsfield unit values. Slow acquisition times also make CBCT vulnerable to motion artifacts from breathing and heartbeat. Because dose calculations in radiotherapy depend on converting Hounsfield units into electron density, international guidelines from EFOMP, ESTRO, and the IAEA recommend that deviations stay within plus or minus 50 HU of baseline values for dose planning purposes. Deviations of about 20 HU in soft tissue and 50 HU in lung and bone generally keep dose calculation errors below one percent, which is why HU accuracy sits at the heart of the evaluation.

The phantom measurements, performed at 100 kV and 125 kV with all available clinical protocols and reconstruction algorithms, revealed a clear density-dependent pattern. For lower-density materials such as air, polymethylpentene, and polyethylene, the standard CBCT actually performed slightly better, while for denser inserts including acrylic, Delrin, and Teflon, HyperSight tracked the CT reference values more closely. The standard system exceeded the 50 HU tolerance for the densest materials in some protocol combinations, whereas HyperSight deviations generally remained within limits. The researchers attribute the improved linearity to the new cesium iodide scintillator, which converts X-rays to signal more efficiently and responds more consistently across a wide density range, a property that matters enormously when electron density maps feed directly into dose calculations.

Uniformity, which measures how consistently pixel values behave across a homogeneous region, also favored the upgraded system, particularly at 100 kV where the standard TrueBeam failed to meet the manufacturer’s plus or minus 30 HU tolerance in some acquisitions. Contrast-to-noise ratio showed a trend toward improvement with HyperSight, with statistically significant differences appearing only in a few specific combinations of tube voltage and reconstruction algorithm. Spatial resolution, quantified through the modulation transfer function, produced mixed results: the larger field of view of HyperSight slightly increases voxel size, which can shave away fine detail, but at the higher energy setting the upgraded system compensated and even outperformed the standard configuration at some contrast levels.

Noise characteristics told a similar story of subtle rather than transformative change. With the older filtered back-projection reconstruction, HyperSight actually showed higher average noise power, though this difference was driven largely by one specific machine, hinting at unit-to-unit variability. With iterative reconstruction algorithms, the noise spectra of the two systems became nearly indistinguishable, suggesting that modern software can suppress machine-dependent noise signatures. Low-contrast visibility, the ability to detect faint differences in tissue attenuation, showed no meaningful difference between the two techniques at either energy, and both systems passed the relevant tolerance criteria in almost all cases.

To test whether the phantom findings translate to the clinic, the team enlisted eight radiographers, the professionals who use CBCT images daily for patient positioning, in a blinded evaluation of scans from 15 patients imaged before and after the HyperSight installation. The evaluators did not know which system produced which image, and volumes were cropped to match fields of view to preserve blinding. The overall result hovered at grade C, meaning no clear preference in image quality or registration accuracy. Yet when the cases were broken down by anatomical region, the radiographers preferred HyperSight in every category except the body region, which included esophageal and femoral cases. The researchers note that status quo bias, the human tendency to favor familiar technology, may have tempered enthusiasm for the new system.

The statistical architecture of the study deserves attention because it illustrates how imaging claims should be scrutinized. The team used pairwise linear mixed models to compare matched acquisitions while accounting for machine-specific and longitudinal variability, and separately applied McNemar’s tests to determine whether differences crossed clinically meaningful tolerance thresholds. Strikingly, the mixed models revealed several differences with large effect sizes, but the McNemar tests confirmed a significant difference in tolerance compliance for only one parameter, Hounsfield unit accuracy, at one measurement point. In other words, the two systems perform similarly within the framework of international guidelines, and the authors argue that the growing importance of adaptive radiotherapy demands more robust and widely accepted image quality tolerance criteria than currently exist.

The study is not without limitations, and the authors are candid about them. Only two machines per group were evaluated at a single center, measurements were limited to three imaging sessions, and the number of paired image slices in each statistical comparison was small, in some cases as few as six. The metal artifact reduction algorithm was not assessed because it is unavailable on the standard system, and the faster acquisition speed of HyperSight, which halves scan time and should reduce motion artifacts, was not directly investigated. The blinded subjective evaluation also required cropping images, which masked one of HyperSight’s headline advantages, its much larger field of view, potentially underestimating its clinical benefit.

What emerges is a portrait of an incremental rather than revolutionary upgrade. HyperSight on C-arm linear accelerators met international and manufacturer tolerance criteria more consistently than the standard system, with the most reliable gains in the Hounsfield unit accuracy that matters most for dose calculation and offline adaptive radiotherapy. Because offline adaptation uses existing clinical tools rather than the full online team of physicists, radiographers, and oncologists, even modest improvements in image fidelity could make adaptive workflows more scalable for clinics worldwide. The Finnish team concludes that patient-based studies across multiple centers are now needed to determine whether the phantom-bench advantages translate into better treatment plans and, ultimately, better outcomes for people with cancer.

Subject of Research: Image quality evaluation of HyperSight cone beam CT on Varian TrueBeam linear accelerators for adaptive radiotherapy

Article Title: Evaluating Image Quality of Cone Beam Computed Tomography Using HyperSight® on the Varian TrueBeam

Article References: Lohela, J., Rytky, S. J. O., Karhula, S. S., & Nikkinen, J. (2026). Evaluating Image Quality of Cone Beam Computed Tomography Using HyperSight® on the Varian TrueBeam. Annals of Biomedical Engineering. https://doi.org/10.1007/s10439-026-04372-7

Image Credits: AI Generated

DOI: 10.1007/s10439-026-04372-7

Keywords: CBCT, HyperSight, TrueBeam, image quality, adaptive radiotherapy, Hounsfield units, radiotherapy, medical imaging, linear accelerator, phantom study, image-guided radiotherapy, dose calculation