5. Radiation Biology and Safety

The dosimetric description of CT begins not with the patient but with the photon fluence emerging from the bowtie-filtered x-ray beam, and proceeds through a hierarchy of physical quantities that must be kept conceptually distinct. The most fundamental is the absorbed dose $D$, the expectation value of energy $\mathrm{d}\bar{\varepsilon}$ imparted by ionizing radiation to matter of mass $\mathrm{d}m$:

$D = \frac{\mathrm{d}\bar{\varepsilon}}{\mathrm{d}m},\qquad [\,\mathrm{Gy}\,] = \mathrm{J\,kg^{-1}}.$

Absorbed dose is a point quantity defined for any material, and it is the physically real determinant of biological injury. It is operationally derived from kerma (kinetic energy released per unit mass), the energy transferred to charged particles per unit mass of the medium, $K = \mathrm{d}E_{tr}/\mathrm{d}m$. Under charged-particle equilibrium and with radiative losses negligible at diagnostic energies (40–140 keV effective), $D \approx K_{col}$, which is why air-kerma measurements in a phantom can be converted to tissue dose through the ratio of mass energy-absorption coefficients $(\mu_{en}/\rho)_{tissue}/(\mu_{en}/\rho)_{air}$. In CT the practical surrogate is the CT dose index, specifically $\mathrm{CTDI}_{vol}$, which folds the helical pitch $p$ into the weighted index $\mathrm{CTDI}_{w}=\tfrac{1}{3}\mathrm{CTDI}_{100,center}+\tfrac{2}{3}\mathrm{CTDI}_{100,periphery}$ via $\mathrm{CTDI}_{vol}=\mathrm{CTDI}_{w}/p$. The dose-length product, $\mathrm{DLP}=\mathrm{CTDI}_{vol}\times L$, integrates over the scanned length $L$ and is the quantity most directly tied to integral energy deposition and hence to stochastic risk. It is essential to internalize that $\mathrm{CTDI}_{vol}$ is a property of the scanner and a reference phantom (16 or 32 cm PMMA), not of the patient; it systematically misrepresents dose in bodies much smaller or larger than the phantom, a discrepancy formalized later by the size-specific dose estimate.

Because different tissues differ in radiosensitivity, and because radiations of differing linear energy transfer (LET) produce different biological effect per gray, absorbed dose alone cannot index detriment. The equivalent dose to an organ, $H_T=\sum_R w_R D_{T,R}$, applies the radiation-weighting factor $w_R$ (unity for the photons and electrons of CT, but 20 for alpha particles), yielding units of sieverts. The effective dose, the most widely quoted and most widely misused quantity in clinical radiology, sums the equivalent doses across organs weighted by the ICRP tissue-weighting factors $w_T$:

$E=\sum_T w_T H_T=\sum_T w_T \sum_R w_R D_{T,R}.$

The $w_T$ values of ICRP Publication 103 (2007) — 0.12 for lung, stomach, colon, breast, red marrow; 0.08 for gonads; 0.04 for bladder, esophagus, liver, thyroid; 0.01 for skin, bone surface, brain, salivary glands — were derived from sex- and age-averaged stochastic detriment in a reference population. This origin dictates the central caveat of expert practice: effective dose is a protection quantity designed for comparing modalities and managing populations, not a patient-specific risk metric. Applying a 70 kg reference $w_T$ schema to a neonate, or quoting $E$ to three significant figures for an individual, is a category error. In practice $E$ is estimated from DLP through region-specific conversion coefficients $k$ (mSv·mGy⁻¹·cm⁻¹): roughly 0.0021 for the head, 0.014 for the chest, and 0.015 for the abdomen-pelvis in adults, with pediatric $k$ values several-fold higher because the same DLP irradiates a smaller, more radiosensitive volume. The failure mode to guard against is treating these conversions as exact; they carry uncertainties of 20–40 percent and are invalid outside the body region and age for which they were tabulated.

The biological consequences of CT exposure are entirely downstream of a stochastic physical event: the deposition of energy in or near the DNA double helix. At diagnostic photon energies the dominant initiating interactions are the photoelectric effect and Compton scattering, both of which liberate secondary electrons whose track structure determines the spatial pattern of ionization. This is the crux of microdosimetry: although CT delivers low-LET radiation in the macroscopic average, the secondary electrons produce dense clusters of ionization at the ends of their tracks, and it is these clusters that generate the biologically critical DNA double-strand break (DSB) and, more importantly, the clustered or complex lesion in which two or more breaks and base damages occur within one to two helical turns. A single gray deposits on the order of $10^5$ ionizations per cell nucleus, producing roughly 1000 single-strand breaks, about 40 DSBs, and a few thousand base lesions. The single-strand breaks and isolated base damages are repaired with high fidelity and are biologically near-silent; the clustered DSB is the lesion that matters, because its complexity defeats faithful repair.

The cell disposes of DSBs through two principal pathways whose competition governs outcome. Non-homologous end joining (NHEJ), active throughout the cell cycle and dominant in G1, directly ligates broken ends through the Ku70/80–DNA-PKcs–LIG4 axis and is fast but error-prone, frequently producing small insertions, deletions, and chromosomal translocations. Homologous recombination (HR), restricted to S and G2 where a sister chromatid is available as template, is high-fidelity but slow. The mutagenic and carcinogenic potential of a given exposure therefore depends not only on dose but on the proliferative state of the tissue — a mechanistic basis for the empirical radiosensitivity of marrow, gut crypt, and gonad. Misrepair generates point mutations and, through illegitimate joining of breaks on different chromosomes, the reciprocal translocations that activate oncogenes or disrupt tumor suppressors. The dicentric chromosome, the cytogenetic signature exploited in biological dosimetry, is a direct readout of this misrepair and rises essentially linearly with dose in the diagnostic-to-low-therapeutic range.

From these mechanisms the two classes of clinical effect diverge sharply. Deterministic effects (now termed tissue reactions) arise when cell killing exceeds the tissue's reserve capacity for repopulation; they exhibit a practical threshold below which the reaction does not clinically manifest, and a severity that escalates with dose above it. Representative thresholds — approximately 0.5 Gy for detectable lens opacity (revised downward by ICRP from the historical 2 Gy as evidence for a much lower or absent cataract threshold accumulated), 2 Gy for transient erythema, 3–6 Gy for temporary epilation, and 0.1 Gy for measurable but reversible suppression of spermatogenesis — are essentially never reached by a single diagnostic CT but are clinically pertinent to CT perfusion, repeated interventional CT, and overlapping multiphase protocols, where fixed-table skin dose can accumulate toward erythema thresholds. Stochastic effects — carcinogenesis and heritable mutation — have, in the prevailing regulatory model, no threshold: the probability (not severity) of the effect scales with dose. The linear-no-threshold (LNT) model, anchored in the atomic-bomb Life Span Study and supported in the diagnostic range by the Pearce (2012, Lancet) and Mathews (2013, BMJ) pediatric CT cohorts, posits excess relative risk proportional to organ dose. The expert must hold this model critically: at the per-examination doses of CT (typically 1–20 mSv effective), the predicted individual lifetime risk increment is small (order $10^{-4}$ to $10^{-3}$) and statistically unverifiable against a baseline lifetime cancer incidence near 40 percent, while reverse causation and confounding by indication threaten the pediatric cohorts. Multiplying such per-capita risks across large populations to compute 'collective dose' attributable deaths is, in the ICRP's own current position, inappropriate; the defensible use of LNT is prospective optimization, not retrospective body-counting.

Translating radiobiology into practice is an exercise in optimization under the ALARA (as low as reasonably achievable) principle, constrained always by the overriding requirement that the study answer its clinical question — an underexposed, non-diagnostic scan delivers dose for zero benefit and is the most common failure of mis-applied dose reduction. The governing relationship is that image noise (the pixel standard deviation $\sigma$) scales inversely with the square root of the photons reaching the detector, so that for fixed reconstruction $\sigma \propto 1/\sqrt{D}$; halving noise costs a quadrupling of dose. Every dose-reduction lever is a negotiation against this relation. Tube-current modulation (automatic exposure control) varies mA angularly and along the z-axis to hold a target noise index constant as patient attenuation changes, typically saving 20–40 percent. Tube-potential (kV) optimization is more physically subtle and more powerful for iodinated studies: lowering kV from 120 to 100 or 80 shifts the spectrum toward the iodine K-edge (33.2 keV), sharply increasing iodine attenuation through the photoelectric effect (whose cross-section scales roughly as $Z^3/E^3$) so that contrast-to-noise ratio per unit dose rises even as raw noise increases — the basis of low-kV CTA. Iterative and deep-learning reconstruction break the historical filtered-back-projection coupling between dose and noise by modeling photon statistics and system optics, permitting 30–70 percent dose reduction at matched noise, though the expert must watch for the characteristic 'plastic' or 'waxy' texture and the potential loss of low-contrast detectability that can mask subtle lesions. Beyond these, restricting scan length to the indicated anatomy (recall $\mathrm{DLP}=\mathrm{CTDI}_{vol}\times L$), eliminating reflexive multiphase acquisitions, and applying organ-based current reduction or bismuth shielding for superficial radiosensitive organs complete the toolkit. The pervasive cognitive failure mode here is dose creep: the gradual upward drift of technique to obtain ever-smoother images, untethered from diagnostic necessity, which dose-monitoring registries and diagnostic reference levels exist to police.

Pediatric considerations elevate optimization from good practice to imperative, for two compounding reasons grounded in the preceding biology. First, children possess more dividing, radiosensitive stem cells and a longer post-exposure lifespan over which a stochastic clone can express, so age-averaged risk coefficients understate their true per-mGy detriment; the Pearce and Mathews cohorts quantify excess leukemia and brain-tumor and solid-cancer incidence after childhood CT, with cumulative doses of 50–60 mGy associated with roughly two- to three-fold relative risks. Second, a small body irradiated to the same $\mathrm{CTDI}_{vol}$ absorbs a far higher organ dose than the 32 cm reference phantom implies — the rationale for the size-specific dose estimate (SSDE), $\mathrm{SSDE}=f_{size}\times\mathrm{CTDI}_{vol}$, where the conversion factor $f_{size}$ derived from effective diameter can exceed 2 in neonates. Practice therefore demands weight- or diameter-based protocol scaling of mA and kV, scrupulous justification under the Image Gently framework, and aggressive collimation, with the explicit recognition that conversion coefficients $k$ for effective dose are several-fold higher in infants. Pregnancy considerations require the physician to reason on two distinct axes. Deterministically, the conceptus is at risk of malformation, growth restriction, and reduced IQ chiefly during organogenesis and early fetal life, but the relevant ICRP and AAPM threshold for these tissue reactions is approximately 100 mGy of conceptus dose — far above the typical fetal dose of a single body CT (commonly 1.5–25 mGy for abdominopelvic acquisitions, and sub-mGy for chest CT or pulmonary embolism studies with appropriate technique). The corollary, endorsed in the AAPM position statement, is that no individual diagnostic CT delivers a fetal dose warranting therapeutic abortion. Stochastically, however, the LNT model still applies, with a per-mGy childhood-cancer risk to the conceptus on the order of $0.006\%$ per mGy; this small absolute increment justifies dose minimization (low-kV, reduced length, shielding where it does not compromise the study) and, where an equivalent non-ionizing test such as ultrasound or MRI answers the question, substitution — but it must never be allowed to deny a pregnant patient an indicated, potentially life-saving CT, as in suspected pulmonary embolism or major trauma. Defensible counseling pairs an explicit conceptus-dose estimate against the 100 mGy threshold with a candid statement of the small stochastic risk relative to the much larger maternal and fetal hazard of a missed diagnosis.