Treatment Planning System (TPS) QA & Patient Verification

Treatment Planning System (TPS) QA & Patient Verification

The treatment planning system (TPS) serves as the computational core of modern radiation oncology. It synthesizes volumetric patient anatomical data from computed tomography (CT), magnetic resonance imaging (MRI), and positron emission tomography (PET) to model complex physical interactions of ionizing radiation within inhomogeneous human tissues. Because errors introduced during TPS commissioning or routine operation propagate systematically to every patient treated under that model, rigorous validation and verification protocols are vital to medical health physics practice. The American Association of Physicists in Medicine (AAPM) provides foundational guidelines through Task Group 53 (TG-53) and Medical Physics Practice Guideline (MPPG) 5.b (2022, which superseded MPPG 5.a), which standardize beam modeling, CT calibration, secondary monitor unit checks, and patient-specific delivery verification.

---

AAPM TG-53 & MPPG 5.b Commissioning and QA Guidelines

Commissioning a photon beam model in a TPS involves acquiring a comprehensive experimental dataset in a water phantom and tuning algorithmic parameters (e.g., source size, electron contamination, multileaf collimator transmission, and head scatter factors) to replicate measured dosimetry.

Core Beam Modeling Components

1. Percentage Depth Dose (PDD) and Tissue Maximum Ratios (TMR): Measured on the central axis across square and rectangular fields ranging from the smallest deliverable field (e.g., $2 \times 2\text{ cm}^2$ or $3 \times 3\text{ cm}^2$) to the maximum aperture ($40 \times 40\text{ cm}^2$). The model must correctly replicate the surface build-up region, depth of dose maximum ($d_{\text{max}}$), and downstream exponential attenuation.
2. Cross-Beam Profiles (In-Plane and Cross-Plane): Acquired at multiple clinically relevant depths—including $d_{\text{max}}$, $5\text{ cm}$, $10\text{ cm}$, $20\text{ cm}$, and $30\text{ cm}$. Profiles validate the modeling of primary horn flattening, field penumbra ($80\%\text{--}20\%$ falloff), and out-of-field low-dose scatter tails.
3. Diagonal Profiles: Acquired across the widest field diagonals (e.g., $40 \times 40\text{ cm}^2$ corners at $56.6\text{ cm}$ diagonal distance) to parameterize circular flattening filter symmetry and off-axis spectral softening.
4. Physical and Enhanced Dynamic Wedges (EDW): Commissioning requires verifying wedge transmission factors across all available wedge angles ($15^\circ, 30^\circ, 45^\circ, 60^\circ$) and field sizes, alongside profile tilt validation resulting from segmented jaw motion.
5. Multileaf Collimator (MLC) Parameters:
   • Average Leaf Transmission: Average of interleaf leakage (between adjacent leaves) and intraleaf transmission (through the solid tungsten body). Typical double-focused or rounded-end MLCs exhibit transmission between $1.0\%$ and $2.0\%$.
   • Dosimetric Leaf Gap (DLG) / Rounded Leaf-End Modeling: Because curved leaf ends produce a moving transmission penumbra, radiation field size differs slightly from the light-projected or mechanical leaf position. The DLG (typically $0.5\text{ to } 2.5\text{ mm}$) accounts for this offset and is critical for accurate dynamic IMRT and VMAT fluence calculations.
   • Leaf Speed and Acceleration Limits: Maximum leaf speeds (a few centimetres per second, depending on the MLC model) must be respected by the optimizer to avoid undeliverable plans.
6. Distance Dependence & Couch Attenuation: Commissioning verifies field divergence and depth-dose changes at shortened and extended SSDs (MPPG 5.b Test 5.7) and includes couch modeling. AAPM TG-176 notes that carbon-fiber couch tops attenuate a few percent of the beam at normal incidence and considerably more at oblique angles through rails or thick sections, so omitting the couch from the TPS model can cause dose errors of several percent.

MPPG 5.b Dosimetric Validation Tolerances

MPPG 5.b sets validation tolerances consistent with those used by IROC Houston:

MPPG 5.b recommends two external end-to-end tests for IMRT/VMAT, such as the IROC Houston head-and-neck credentialing phantom and a heterogeneous thoracic phantom.

CT-to-Electron Density Calibration & Artifact Corrections

Dose calculation algorithms (such as Collapsed Cone Convolution Superposition, Anisotropic Analytical Algorithm [AAA], and Monte Carlo) compute photon transport and energy deposition in heterogeneous media by scaling physical photon interaction cross sections according to the medium's relative electron density ($\rho_e$, normalized to liquid water at $4^\circ\text{C}$ where $\rho_{e,\text{water}} = 1.000$).

The Hounsfield Unit (HU) Calibration Curve

Computed tomography scanners measure linear attenuation coefficients ($\bar{\mu}$) of diagnostic X-ray spectra ($80\text{ to } 140\text{ kVp}$), mapping each volumetric voxel to a calibrated CT number in Hounsfield Units (HU):

$\text{HU} = 1000 \times \left(\frac{\bar{\mu}_{\text{tissue}} - \bar{\mu}_{\text{water}}}{\bar{\mu}_{\text{water}}}\right)$

By definition, air possesses $\text{HU} = -1000$ and water possesses $\text{HU} = 0$. In clinical soft tissues, the relationship between HU and relative electron density is approximately linear, but it branches into distinct linear segments across the full physical range:

Relative Electron Density (ρe)
    ^
2.0 |                                                / (Cortical Bone: HU ≈ +1800, ρe ≈ 1.85)
    |                                              /
1.5 |                                    / (Trabecular Bone: HU ≈ +300, ρe ≈ 1.15)
    |                         / (Soft Tissue / Water: HU = 0, ρe = 1.00)
1.0 |            / (Adipose: HU ≈ -100, ρe ≈ 0.95)
    |          /
0.5 |    / (Inhaled Lung: HU ≈ -700, ρe ≈ 0.30)
    |  /
0.0 +----------------------------------------------------> CT Number (HU)
 -1000        -500          0          +500       +1000      +1500      +2000

Stoichiometric Calibration Methodology

Direct empirical calibration using tissue-substitute plastic inserts (e.g., epoxy resins doped with calcium compounds) introduces systematic errors because plastic additives do not replicate the exact atomic composition ($Z_{\text{eff}}$) of human tissues. Under the stoichiometric calibration method (developed by Schneider, Bortfeld, and Schlegel):

1. Measured CT numbers of certified tissue-equivalent phantom plugs are acquired on the clinical CT scanner.
2. A theoretical parameterization models the scanner's effective energy by separating photoelectric, Compton, and coherent scattering components.
3. The parameterized model computes predicted HUs for standard human tissues (ICRU Report 44/46 composition).
4. A piecewise bilinear or multilinear calibration curve is generated for clinical TPS configuration.

Clinical Artifacts: Contrast Media and Metal Implants

Clinical CT datasets frequently present non-physiological attenuation anomalies that, if left uncorrected, induce severe therapeutic dose computation errors:

• Intravenous (IV) and Oral Contrast Media: Contrast agents contain high atomic number elements such as iodine ($Z = 53$) or barium ($Z = 56$). At diagnostic CT energies ($80\text{--}140\text{ kVp}$), the photoelectric effect dominates (cross section $\tau \propto Z^3 / E^3$), so contrast-filled vessels and bowel show CT numbers raised by a few hundred HU or more (dense oral barium can exceed $+1000\text{ HU}$). Therapeutic megavoltage beams ($6\text{--}18\text{ MV}$) interact mainly by Compton scattering, which depends on electron density rather than $Z$, so the contrast adds little real attenuation. If the uncorrected CT dataset is used, the TPS reads these voxels as denser tissue, overestimates the attenuation, and assigns extra monitor units, overdosing tissue downstream. For intravenous contrast the effect is usually small (a few percent or less), but large dense regions such as a bowel full of oral contrast can cause larger errors. Common policies are to plan on a non-contrast scan or to contour contrast regions and override them to an appropriate tissue density.
• Metallic Implants (Hip Prostheses, Spinal Fixation Rods, Dental Gold): High-$Z$, high-density metals (titanium $\rho = 4.5\text{ g/cm}^3$, stainless steel $\rho = 8.0\text{ g/cm}^3$, cobalt-chromium $\rho = 8.5\text{ g/cm}^3$) completely attenuate diagnostic X-ray beams, generating severe photon starvation, beam hardening streaks, and zero-count dark bands across the pelvis or head and neck. Metal Artifact Reduction (MAR) reconstruction algorithms mitigate visual streaking. For dose computation, medical physicists must manually contour the metallic foreign bodies, assign certified physical densities (e.g., $\rho = 4.54\text{ g/cm}^3$, $\rho_e = 3.73$ for titanium alloy), override artificial dark streak bands back to normal soft tissue ($\rho_e = 1.0$), and orient treatment beam angles to completely avoid traversing the metallic prosthesis.

Independent Secondary Monitor Unit (MU) Verification

Independent secondary verification of monitor units (or point dose) constitutes an essential quality control barrier in external beam radiotherapy, designed to catch transcription errors, incorrect algorithmic normalization, improper heterogeneity handling, or corrupted beam models before the first patient fraction.

Analytical MU Formulation (3D Conformal Radiotherapy)

For an isocentric (SAD) treatment setup delivering an absorbed dose $D$ at depth $d$ to a target volume, the monitor units are computed via classical dosimetric formalisms:

$\text{MU} = \frac{D}{D'_0 \times S_c(r_c) \times S_p(r_p) \times \text{TMR}(d, r_d) \times \text{OAR}(x,y) \times \text{WF} \times \text{TF} \times \left(\frac{\text{SAD}}{\text{SPD}}\right)^2}$

where:

• $D'_0$ is the reference output for a $10 \times 10\text{ cm}^2$ field at $d_{\text{max}}$ at the reference distance used with TMRs. If the machine is calibrated to $1.000\text{ cGy/MU}$ at $d_{\text{max}}$ for $100\text{ cm}$ SSD, the output at an isocenter $100\text{ cm}$ from the source is $1.000 \times \left(\frac{100 + d_{\text{max}}}{100}\right)^2$, about $1.030\text{ cGy/MU}$ for $6\text{ MV}$ ($d_{\text{max}} = 1.5\text{ cm}$).
• $S_c(r_c)$ is the collimator scatter factor, measured in air with a mini-phantom as a function of jaw opening $r_c$.
• $S_p(r_p)$ is the phantom scatter factor, reflecting scatter generated inside the patient as a function of the blocked/collimated field size $r_p$ at the patient surface.
• $\text{TMR}(d, r_d)$ is the Tissue-Maximum Ratio at depth $d$ for equivalent field size $r_d$ at depth.
• $\text{OAR}(x,y)$ is the off-axis ratio accounting for beam profile horns and flattening filter radial variation.
• $\text{WF}$ and $\text{TF}$ are the transmission factors for physical/dynamic wedges and tray accessories.
• $\left(\frac{\text{SAD}}{\text{SPD}}\right)^2$ is the inverse-square correction from the distance at which $D'_0$ is defined (here the $100\text{ cm}$ isocenter) to the source-to-calculation-point distance ($\text{SPD}$). With the point at isocenter, this factor equals $1.00$.

Worked Clinical MU Calculation

A radiation oncologist prescribes $200\text{ cGy}$ to the isocenter of an open $6\text{ MV}$ anterior photon field. The target is located at an effective radiological depth $d_{\text{eff}} = 10.0\text{ cm}$ in an isocentric setup ($\text{SAD} = 100.0\text{ cm}$, $\text{SPD} = 100.0\text{ cm}$). Collimator jaw settings define a field size of $14 \times 18\text{ cm}^2$.

Step 1: Calculate Equivalent Square Field Size

Using Sterling's rule for a rectangular field $A \times B$:

$r_{\text{eq}} = \frac{4 \times \text{Area}}{\text{Perimeter}} = \frac{4 \times (14 \times 18)}{2 \times (14 + 18)} = \frac{1,008}{64} = 15.75\text{ cm}$

Step 2: Retrieve Calibrated Dosimetric Factors

From commissioned clinical tables at $r_{\text{eq}} = 15.75\text{ cm}$:

• $D'_0 = 1.000\text{ cGy/MU}$ (this machine is calibrated to give $1.000\text{ cGy/MU}$ at $d_{\text{max}}$ at the $100\text{ cm}$ isocenter)
• $S_c(15.75) = 1.021$
• $S_p(15.75) = 1.014$
• $\text{TMR}(10.0\text{ cm}, 15.75\text{ cm}) = 0.818$
• $\text{OAR}(0,0) = 1.000$ (on central axis)
• $\text{WF} = 1.000$ (open field), $\text{TF} = 1.000$ (open beam)

Step 3: Compute Monitor Units

$\text{Total Output Factor} = 1.000 \times 1.021 \times 1.014 \times 0.818 = 0.8469\text{ cGy/MU}$

$\text{MU} = \frac{200\text{ cGy}}{0.8469\text{ cGy/MU}} = 236.15 \approx 236\text{ MU}$

Secondary 3D Dose Engines & Action Tolerances

For modulated delivery (IMRT and VMAT), simple hand calculations cannot model MLC motion, variable dose rate, and gantry rotation, so clinics use independent secondary dose engines (separate convolution or Monte Carlo codes) as described in AAPM TG-219. Typical action levels:

• Conventional fields (AAPM TG-114): Suggested action levels range from about $2\%$ for simple fields in homogeneous geometry to about $3\%$–$5\%$ for blocked, wedged, off-axis, or heterogeneous geometries. A difference above the action level is investigated and resolved before treatment.
• IMRT / VMAT: Each clinic sets action levels for its secondary-calculation software and techniques; when the difference exceeds them, the physicist withholds approval and checks density assignments, heterogeneity handling, small-field modeling, and the point's location relative to dose gradients.

Patient-Specific IMRT / VMAT QA & Gamma Index Evaluation

While independent MU engines verify calculation integrity, physical deliverability requires pre-treatment experimental verification. Highly modulated IMRT/VMAT fields feature tight leaf apertures, tongue-and-groove leaf interactions, and rapid gantry acceleration that push linear accelerator hardware to mechanical limits.

Measurement Modalities

• 2D Diode Arrays (e.g., Sun Nuclear MapCHECK): High intrinsic sensitivity and spatial resolution ($7\text{ mm}$ spacing, sub-millimeter detector size). Fast readout, but solid-state diodes exhibit directional dependence and energy dependence to scattered low-energy photons.
• 2D Ionization Chamber Arrays (e.g., PTW Octavius / MatriXX): Outstanding dose linearity and minimal energy dependence, but larger physical chamber cavities ($0.05\text{ to } 0.1\text{ cm}^3$) introduce partial volume averaging in high-gradient penumbras.
• Electronic Portal Imaging Devices (EPID): Amorphous silicon (a-Si) panels offer sub-millimeter spatial resolution ($0.4\text{ mm}$ pixel pitch). Pre-treatment portal dosimetry measures integrated transit or fluence maps with high efficiency.
• 3D Helical & Cylindrical Arrays (e.g., ArcCHECK, Delta4): Continuous cylindrical geometry captures full $360^\circ$ rotational VMAT delivery without angular cosine falloff.

Low's Gamma ($\gamma$) Index Formulation

Comparing a 2D calculated dose distribution $D_c(\vec{r}_c)$ against a measured dose distribution $D_m(\vec{r}_m)$ requires a unified metric that handles both flat dose plateaus (where small dose errors dominate) and steep dose penumbras (where microscopic positional offsets mimic massive percentage errors). Low et al. (1998) introduced the Gamma Index ($\gamma$).

For each measured point $\vec{r}_m$, the generalized distance-dose metric $\Gamma(\vec{r}_m, \vec{r}_c)$ is evaluated relative to every calculated point $\vec{r}_c$:

$\Gamma(\vec{r}_m, \vec{r}_c) = \sqrt{\left(\frac{|\vec{r}_m - \vec{r}_c|}{\Delta d_{\text{max}}}\right)^2 + \left(\frac{D_m(\vec{r}_m) - D_c(\vec{r}_c)}{\Delta D_{\text{max}}}\right)^2}$

The gamma value $\gamma(\vec{r}_m)$ at measured position $\vec{r}_m$ is defined as the minimum value across all candidate calculation points:

$\gamma(\vec{r}_m) = \min_{\{\vec{r}_c\}} \left\{ \Gamma(\vec{r}_m, \vec{r}_c) \right\}$

• $\gamma(\vec{r}_m) \le 1.0$: Point passed (measurement falls within acceptance ellipsoid).
• $\gamma(\vec{r}_m) > 1.0$: Point failed (discrepancy exceeds acceptance criteria).

where $\Delta d_{\text{max}}$ is the Distance-to-Agreement (DTA) acceptance criterion (typically $2\text{ or } 3\text{ mm}$), and $\Delta D_{\text{max}}$ is the Dose Difference (DD) criterion (typically $2\%\text{ or } 3\%$).

Global vs. Local Normalization

The choice of dose normalization in computing $\Delta D_{\text{max}}$ profoundly impacts gamma test sensitivity:

• Global Normalization: The dose difference criterion is scaled relative to a single universal reference value across the entire field—typically the maximum planned dose ($D_{\text{max}}$) or the prescription dose ($D_{\text{rx}}$): $\Delta D_{\text{max}} = \delta \times D_{\text{max}}$ For a $3\%$ global criterion with $D_{\text{max}} = 200\text{ cGy}$, the absolute dose difference tolerance is universally fixed at $\Delta D_{\text{max}} = 0.03 \times 200 = 6.0\text{ cGy}$ for every single point in the phantom. In peripheral low-dose regions (e.g., where dose is $30\text{ cGy}$), an absolute error of $5\text{ cGy}$ is allowed to pass, even though it represents a massive $16.7\%$ relative error. Consequently, global gamma analysis is forgiving and can obscure clinically significant overdoses to adjacent organs-at-risk.
• Local Normalization: The dose difference criterion is evaluated relative to the local calculated dose at that specific point: $\Delta D_{\text{max}}(\vec{r}_c) = \delta \times D_c(\vec{r}_c)$ At a $30\text{ cGy}$ point, a $3\%$ local criterion allows only $\Delta D_{\text{max}} = 0.03 \times 30 = 0.9\text{ cGy}$. A $5\text{ cGy}$ discrepancy yields $\gamma \approx 5.6$, a clear failure. Local normalization is mathematically rigorous and highly sensitive to low-dose and out-of-field modeling errors.

AAPM TG-218 National Consensus Standards

To reduce variability in clinical practice, AAPM Task Group 218 made these recommendations for patient-specific IMRT/VMAT QA:

TG-218 also recommends using statistical process control to set institution-specific tolerance and action limits, and true composite (all fields delivered to the phantom as planned) or field-by-field measurement methods. During commissioning, MPPG 5.b suggests reviewing a stricter $2\%/2\text{ mm}$ analysis (including local normalization) to expose problems hidden by high global pass rates.

Low-Dose Thresholding: The $10\%$ threshold excludes low-dose points outside the beam aperture; including them would inflate the pass rate under global normalization.