Selecting beam delivery angles under proton range uncertainty
In photon therapy, a beam's dose deposition is largely governed by exponential attenuation — angle choice mainly trades off which tissues receive low-dose entrance/exit dose, and modern IMRT/VMAT can average this out over many angles. Protons behave completely differently: they stop. Where they stop — the Bragg peak — depends sensitively on exactly what tissue the beam traversed to get there, making the entry angle a first-order determinant of treatment accuracy, not a minor optimization knob.
A proton beam has a finite, energy-defined range: it deposits a low, roughly flat dose along its entry path, then a sharp spike of dose — the Bragg peak — at a depth set by its initial energy and the stopping power of everything it passed through, then essentially zero dose beyond. This is the core rationale for proton therapy: near-zero exit dose.
But that same finite range is a liability. The depth at which the beam stops (the "range") is calculated from a conversion of CT Hounsfield units to relative stopping power (RSP) along the beam path. Any angle-dependent error in that conversion — from tissue composition, motion, or anatomy change — shifts the Bragg peak forward or backward. A photon beam that is 3% wrong in attenuation assumptions barely changes the dose pattern; a proton beam that is 3% wrong in range calculation can place its peak entirely outside the tumor.
Because range uncertainty is a function of the specific tissues traversed, it is inherently angle-dependent — two entry angles into the same tumor can carry very different amounts of risk, even though the tumor itself hasn't moved.
Angle optimization begins with a broad geometric survey: candidate beam directions are sampled around the patient — commonly every 15–45° depending on anatomical complexity — and evaluated against basic feasibility constraints before any dose calculation is performed.
Early-stage screening typically checks: • Whether the beam path is physically deliverable given gantry/couch mechanical limits and patient positioning • Whether it enters through skin/air rather than through a couch rail, immobilization device, or other hardware • Whether it provides a geometrically direct route to the target without traversing an excessive path length
This coarse survey narrows a continuous 360° (or, for non-coplanar planning, a full spherical) search space down to a manageable shortlist for the detailed heterogeneity and OAR-proximity analysis that follows.
Photon IMRT/VMAT plans often use many beam angles (or a continuous arc) specifically because averaging dose contributions from many directions smooths out entrance-dose hot spots and improves conformity — angle choice is somewhat forgiving because no single angle carries disproportionate risk.
Proton plans invert this logic. Because each field's distal falloff must be trusted to land in exactly the right place, adding more beams does not necessarily average away range error — it multiplies the number of directions in which a range miscalculation could cause harm. Clinically, proton plans therefore tend to use fewer, more deliberately chosen fields (commonly 2–4), sometimes just a single well-chosen field such as a posterior-anterior beam for paraspinal or craniospinal targets where the spinal cord sits immediately distal and must never receive an under-ranged overshoot.
The single largest source of proton range error is not any one measurement mistake — it is the beam's path through anatomically heterogeneous tissue. Wherever a beam edge grazes a bone-air, bone-soft-tissue, or air-tissue interface, protons traveling through slightly different lateral positions see very different stopping powers, and the sharp distal falloff that makes protons attractive degrades into a blurred, unreliable edge.
A proton beam is not a single ray — it has finite lateral width, and even a narrow pencil beam spreads via multiple Coulomb scattering as it traverses tissue. If the beam edge crosses a density interface (e.g. the edge of a rib, or the boundary of an air-filled sinus or bowel gas pocket), different parts of the beam's lateral profile traverse different amounts of bone, air, and soft tissue on their way to the target.
Protons that passed mostly through low-density air arrive at a given depth having lost little energy — they overshoot. Protons that passed through dense bone lose energy fast and undershoot. The result is that instead of one sharp Bragg peak, the beam produces a smeared range distribution: the clean, ~1 mm distal falloff achievable in homogeneous water degrades into an edge blurred over several millimeters to more than a centimeter, directly beneath the heterogeneity.
This is why angle selection actively routes beams around bone-air and bone-tissue interfaces wherever geometrically possible, rather than relying on dose calculation algorithms alone to "average out" the effect after the fact.
Heterogeneity-driven range mixing is most severe in anatomically complex regions:
• Lung: tumor surrounded by low-density aerated lung parenchyma, with the beam potentially clipping rib or vertebral bone — among the hardest sites for robust proton planning, and subject to additional range change from breathing motion between planning and treatment • Head and neck: dense skull base bone, air-filled sinuses and mastoid cavities, and tightly packed organs at risk in a small volume • Pelvis (rectum/bladder filling): air and stool content in the rectum changes day to day, altering the stopping power along posterior beam paths
By contrast, simple, homogeneous-tissue sites — a straightforward prostate case surrounded by soft tissue, for example — present a comparatively forgiving angle-selection problem, which is part of why prostate was among the earliest and most standardized proton indications.
Modern treatment planning systems evaluate each candidate angle by computing water-equivalent path length (WEPL) not just along the central axis but across the beam's full lateral extent, flagging angles where WEPL varies sharply between adjacent ray-paths (a proxy for lateral heterogeneity beneath the beam).
Some centers also compute a "heterogeneity index" or use Monte Carlo range calculations — more accurate than analytic pencil-beam algorithms specifically because Monte Carlo correctly models the scattering physics that generates range mixing — to score each candidate angle before committing to a beam arrangement. Angles that produce low lateral WEPL variance are scored favorable; angles that cross rib edges, vertebral bodies, or sinus/bowel-gas boundaries are scored unfavorable and are deprioritized or eliminated from the candidate list.
Even a beam that travels through perfectly homogeneous tissue still carries irreducible range uncertainty from calibration and biological factors. The clinically decisive question becomes: where does that uncertainty land? Placing the sharp, uncertain distal edge of the Bragg peak directly against a critical organ is the single riskiest geometric decision in proton planning — because a small range error there translates directly into an organ overdose or a tumor underdose.
Along the lateral edges of a proton field, dose falls off due to beam penumbra — a geometric, well-characterized, and comparatively stable effect similar in spirit to photon penumbra. Along the distal edge, dose falls off because the protons have physically stopped, and exactly where they stop depends on the cumulative, angle-specific range calculation described in Stage 2.
That means the distal edge is precisely where the plan's most uncertain parameter — beam range — has its full effect. If a serial organ (spinal cord, brainstem, optic chiasm) sits just beyond the tumor along the beam's path, a range overshoot of only a few millimeters — well within normal calculation uncertainty — can push a damaging dose directly into it. The same overshoot occurring laterally, where penumbra is well-modeled and stable, is far less consequential.
The clinical rule of thumb: never intentionally place the distal 90%–10% falloff region of a field immediately adjacent to a serial, dose-limiting OAR if a comparably effective angle exists whose distal edge falls in more forgiving tissue.
Selecting angles by distal-edge placement blends two kinds of criteria:
• Geometric: for each candidate angle, project the beam's central axis and field edges past the distal target boundary and measure the physical distance to the nearest critical OAR surface. Angles with larger clearance are preferred.
• Dosimetric: because clearance alone doesn't capture uncertainty magnitude, planners (or automated multi-criteria optimization algorithms) also weight this by the OAR's dose tolerance and serial vs. parallel architecture — a few millimeters of clearance may be adequate for a parallel organ (where a small volume receiving high dose is tolerable) but inadequate for a serial organ like the cord, where the maximum point dose anywhere along its length is dose-limiting.
Angles passing both checks are retained; angles failing either are down-ranked or excluded from the final beam arrangement.
Careful angle selection reduces risk but cannot eliminate range uncertainty. Modern proton planning therefore pairs angle optimization with robust optimization: instead of optimizing dose for a single nominal range/position, the planning system simultaneously optimizes across many perturbed scenarios — commonly the nominal case plus range shifted ±2.5–3.5% and patient setup shifted ±3–5 mm in each direction (often 9 to 21+ combined scenarios).
A plan is only accepted if target coverage and OAR sparing hold up acceptably across this whole scenario envelope, not just in the ideal case. Good angle selection makes robust optimization's job easier — an angle whose distal edge already sits safely away from an OAR gives the optimizer much more room to absorb worst-case range shifts without violating OAR constraints.
Once candidate angles have been screened for tissue heterogeneity and distal-edge OAR proximity, the planning team locks in a final, small set of beam directions. Compared to photon IMRT's many-angle, highly modulated arrangements, proton plans favor a deliberately sparse set of fields, each chosen because its entry, path, and distal falloff are individually defensible.
The angles surviving heterogeneity and OAR-proximity screening form a shortlist, but the final selection also considers how the beams work together as a set:
• Angular diversity: beams spaced widely enough apart that their individual weaknesses (e.g. a slightly compromised lateral margin) don't compound in the same location • Redundant coverage: with the tumor covered from multiple directions, no single beam's range error alone determines whether the tumor is adequately treated • Range-mixing avoidance across the whole set: a combination of angles is checked jointly, not just individually, since a plan could pass every single-beam check yet still concentrate distal uncertainty in one spot if all beams converge on it from similar directions
Modern proton centers deliver scanned pencil-beam fields using one of two related strategies once angles are fixed:
• SFUD (Single-Field Uniform Dose): each individual field is optimized to deliver a uniform dose to the target on its own. Plan robustness against any one field's range error is easier to reason about, at some cost to normal-tissue sparing efficiency.
• IMPT (Intensity-Modulated Proton Therapy): all fields are optimized simultaneously, and no single field needs to be uniform by itself — the fields' non-uniform contributions sum to the desired target dose. This gives superior OAR sparing and conformity but makes the plan more sensitive to range and setup errors in any one field, which is exactly why robust optimization (Stage 3) is essential for IMPT.
Both approaches depend entirely on the beam angles chosen beforehand — poor angle selection cannot be fully corrected by either optimization method.
A single well-placed posterior field is sometimes the entire plan for a paraspinal tumor: one field, one distal edge to manage, and the spinal cord kept proximal (upstream) of the Bragg peak rather than distal to it — inherently safer than any arrangement placing the cord beyond the peak.
Angle selection is not a purely abstract optimization — it is bounded by what the physical gantry can actually deliver. Isocentric proton gantries are among the largest moving medical devices ever built: cyclotron- or synchrotron-fed beamlines must be bent through large-radius magnets to reach the treatment head from any angle around the patient, producing structures weighing 90–190+ metric tons and standing several stories tall.
This mass makes gantry rotation slow (often one to a few minutes per major angle change) compared to a few seconds for a conventional linear accelerator gantry, which is one reason proton plans favor fewer, carefully chosen static angles over continuous-arc delivery. It has also driven strong recent interest in compact, lighter gantries using superconducting magnets, and in fixed-beamline rooms that instead rotate the patient couch to change the effective beam angle — trading some angular flexibility for dramatically lower cost, size, and delivery time.
The payoff of careful angle selection is visible in the final dose distribution: a dose cloud that conforms tightly to the tumor volume, spares nearby organs at risk, and remains robust to the residual range and setup uncertainty that no amount of planning can fully eliminate. Comparing this to a naive, geometrically convenient but dosimetrically poor angle choice makes the value of the angle-optimization process concrete.
With angles selected to avoid heterogeneous paths and keep distal edges away from OARs, the summed dose from all selected fields builds a high-dose region that closely matches the shape of the tumor: sharp lateral penumbra, and — crucially — a trustworthy distal edge that stops close to where it was planned to stop, because the path used to get there was chosen specifically to minimize range-calculation error.
Organs at risk sitting adjacent to, but outside, the target receive comparatively low dose, both because no beam's distal uncertainty band overlaps them and because proton beams deposit essentially no dose beyond the Bragg peak — unlike photon beams, which continue depositing exit dose through any tissue beyond the target regardless of angle.
A "naive" angle choice — selected purely for a short, geometrically direct path to the tumor, without regard to tissue heterogeneity or distal OAR proximity — can look perfectly reasonable on a simple beam's-eye-view projection while hiding a serious dosimetric flaw: its distal falloff may sit directly against a spinal cord, brainstem, or other serial organ, or its path may cross a rib/air interface that blurs the falloff into a much wider uncertain band than the geometry alone suggests.
Under the same range and setup uncertainty that a well-chosen angle would absorb safely, the naive angle can transmit meaningfully more dose into the OAR — precisely because the very source of proton therapy's advantage (a sharp, finite range) becomes a liability when that range terminates in the wrong place.
This is the central paradox of proton planning: the property that makes protons dosimetrically superior to photons (a hard stop) is also the property that makes angle selection so much more consequential — a photon beam that "misses" its ideal angle degrades gracefully, a proton beam that misses can fail sharply.
Clinically, final angle selection, robust optimization, and rigorous plan review form a single integrated workflow rather than sequential independent steps. Once beams are chosen and optimized, plans are evaluated using robustness analysis tools that recompute dose across the full range/setup scenario envelope (Stage 3) and confirm that target coverage and OAR constraints hold across all of them — not merely in the nominal, best-case scenario.
When a plan fails robustness checks, the response is often to revisit angle selection itself — swap a marginal angle for an alternative from the candidate shortlist, adjust field weighting, or in difficult anatomy accept a compromise plan with tighter OAR margins — before falling back to purely mathematical optimization fixes. Good angle geometry, chosen with an understanding of range physics, remains the foundation that everything downstream depends on.