In radiation therapy, the human body is often simplified as a uniform volume of water to make dose calculations computationally feasible. However, the real body is a complex landscape of sharp density gradients—interfaces where radiation passes from low-density lung tissue to high-density bone, or through air pockets in the digestive tract.
Traditional dose calculation engines struggle at these boundaries. When radiation moves through heterogeneous tissue, the physical behavior of particles changes abruptly, leading to "electronic disequilibrium" and scattering effects that simplified models fail to capture. Research in this field focuses on bridging the gap between slow, "gold-standard" simulations and the fast, approximate models required for clinical practice.
The Interface Problem
The primary challenge in dosimetry is the transition between different materials. As Zheng et al. demonstrate in the context of lung stereotactic body radiotherapy (SBRT), simplified "pencil beam" (PB) algorithms often overestimate the dose delivered to a tumor. This happens because PB models cannot accurately account for the way electrons scatter at the interface between a dense tumor and the surrounding low-density lung tissue.
These inaccuracies are not merely academic; they have direct clinical consequences. Sengupta et al. note that achieving a uniform dose across the entire body is exceptionally difficult because anatomical thickness and tissue density vary wildly between the neck, lungs, and abdomen. If a model ignores these variations—such as failing to map internal air cavities or skeletal structures—it can significantly underestimate the radiation exposure to critical organs like the heart Wang et al..
The Speed-Accuracy Trade-off
Clinicians currently face a binary choice:
- Monte Carlo (MC) Simulations: These track individual particle interactions, providing high accuracy in heterogeneous media, but are computationally expensive and slow.
- Analytical Models: These use mathematical approximations for speed but often rely on assumptions of tissue uniformity that fail at complex anatomical interfaces Bangert et al..
To break this trade-off, researchers are developing "deterministic" mathematical shortcuts. For example, Stammer et al. proposed a high-order "low-rank" method that achieves the accuracy of massive supercomputer simulations on a single workstation by reducing the mathematical complexity of proton transport. Similarly, Bangert et al. introduced Analytical Probabilistic Modeling (APM), which uses closed-form Gaussian equations to calculate dose variance without needing thousands of separate simulations.
Deep Learning as a Physics Proxy
A recent shift in the field involves using deep learning to "learn" the physics of particle transport. Instead of solving differential equations, these models treat dose prediction as a pattern-recognition task.
Pastor-Serrano and Perkó developed the DoTA (Dose Transformer Algorithm), which frames energy deposition as a sequence-modeling task. By training on Monte Carlo data, the model can predict doses 4,000 times faster than traditional simulations while maintaining high precision. To handle the specific challenges of heterogeneity, Zhang et al. introduced a "sliding window" technique that allows neural networks to focus on local density gradients, ensuring that the model remains robust even when transitioning between air, bone, and soft tissue.
Adaptive Modeling and Tissue Classification
For these models to work, the underlying anatomical data must be accurate. In PET/MRI imaging, Martínez-Möller et al. validated a "4-class" segmentation approach—dividing the body into background, lungs, fat, and soft tissue—to correct for photon attenuation. While effective, the authors noted that static density values for lungs fail to account for the vast differences between patients, suggesting a need for more granular, gradient-based mapping.
In modern online adaptive radiotherapy (oART), where plans are adjusted daily based on new scans, the accuracy of "synthetic" images is vital. While Dona Lemus et al. found that current workflows are relatively robust to minor air-mapping errors in the prostate, the field continues to push for better material overrides. For instance, Hooshangnejad et al. showed that treating iodinated spacers as water rather than high-density material prevents unintended dose deviations in proton therapy.
Ultimately, the goal of modeling tissue heterogeneity is to move toward a "physics-inspired" approach that is fast enough for real-time use but granular enough to respect the complex boundaries of human anatomy De Benetti et al..
Key Foundational Work
Effect of skull shape approximations in Gamma Knife dose calculations (2007) — contributed 3 concept elements to this direction.
Tissue Classification as a Potential Approach for Attenuation Correction in Whole-Body PET/MRI: Evaluation with PET/CT Data (2009) — contributed 1 concept elements to this direction.