Artificial intelligence (AI) refers to methods that improve and automate challenging human tasks by systematically capturing and applying relevant knowledge in these tasks. Over the past decades, a number of approaches have been developed to address different types and needs of system intelligence ranging from search strategies to knowledge representation and inference to robotic planning. In the context of radiation treatment planning, multiple AI approaches may be adopted to improve the planning quality and efficiency. For example, knowledge representation and inference methods may improve dose prescription by integrating and reasoning about the domain knowledge described in many clinical guidelines and clinical trials reports. In this review, we will focus on the most studied AI approach in intensity modulated radiation therapy (IMRT)/volumetric modulated arc therapy (VMAT)-machine learning (ML) and describe our recent efforts in applying ML to improve the quality, consistency, and efficiency of IMRT/VMAT planning.With the available high-quality data, we can build models to accurately predict critical variables for each step of the planning process and thus automate and improve its outcomes. Specific to the IMRT/VMAT planning process, we can build models for each of the four critical components in the process: DVH, Dose, Fluence, and Human Planner. These models can be divided into two general groups. The first group focuses on encoding prior experience and knowledge through ML and more recently deep learning (DL) from prior clinical plans and using these models to predict the optimal DVH (DVH prediction model), or 3D dose distribution (dose prediction model), or fluence map (fluence map model). The goal of these models is to reduce or remove the trial-and-error process and guarantee consistently high-quality plans. The second group of models focuses on mimicking human planners' decision-making process (planning strategy model) during the iterative adjustments/guidance of the optimization engine. Each critical step of the IMRT/VMAT treatment planning process can be improved and automated by AI methods. As more training data becomes available and more sophisticated models are developed, we can expect that the AI methods in treatment planning will continue to improve accuracy, efficiency, and robustness.

Purpose. We have previously reported an artificial intelligence (AI) agent that automatically generates intensity-modulated radiation therapy (IMRT) plans via fluence map prediction, by-passing inverse planning. This AI agent achieved clinically comparable quality for prostate cases, but its performance on head-and-neck patients leaves room for improvement. This study aims to collect insights of the deep-learning-based (DL-based) fluence map prediction model by systematically analyzing its prediction errors. Methods. From the modeling perspective, the DL model’s output is the fluence maps of IMRT plans. However, from the clinical planning perspective, the plan quality evaluation should be based on the clinical dosimetric criteria such as dose-volume histograms. To account for the complex and non-intuitive relationships between fluence map prediction errors and the corresponding dose distribution changes, we propose a novel error analysis approach that systematically examines plan dosimetric changes that are induced by varying amounts of fluence prediction errors. We investigated four decomposition modes of model prediction errors. The two spatial domain decompositions are based on fluence intensity and fluence gradient. The two frequency domain decompositions are based on Fourier-space banded frequency rings and Fourier-space truncated low-frequency disks. The decomposed error was analyzed for its impact on the resulting plans’ dosimetric metrics. The analysis was conducted on 15 test cases spared from the 200 training and 16 validation cases used to train the model. Results. Most planning target volume metrics were significantly correlated with most error decompositions. The Fourier space disk radii had the largest Spearman’s coefficients. The low-frequency region within a disk of ∼20% Fourier space contained most of errors that impact overall plan quality. Conclusions. This study demonstrates the feasibility of using fluence map prediction error analysis to understand the AI agent’s performance. Such insights will help fine-tune the DL models in architecture design and loss function selection.

Deep learning algorithms for radiation therapy treatment planning automation require large patient datasets and complex architectures that often take hundreds of hours to train. Some of these algorithms require constant dose updating (such as with reinforcement learning) and may take days. When these algorithms rely on commerical treatment planning systems to perform dose calculations, the data pipeline becomes the bottleneck of the entire algorithm’s efficiency. Further, uniformly accurate distributions are not always needed for the training and approximations can be introduced to speed up the process without affecting the outcome. These approximations not only speed up the calculation process, but allow for custom algorithms to be written specifically for the purposes of use in AI/ML applications where the dose and fluence must be calculated a multitude of times for a multitude of different situations. Here we present and investigate the effect of introducing matrix sparsity through kernel truncation on the dose calculation for the purposes of fluence optimzation within these AI/ML algorithms. The basis for this algorithm relies on voxel discrimination in which numerous voxels are pruned from the computationally expensive part of the calculation. This results in a significant reduction in computation time and storage. Comparing our dose calculation against calculations in both a water phantom and patient anatomy in Eclipse without heterogenity corrections produced gamma index passing rates around 99% for individual and composite beams with uniform fluence and around 98% for beams with a modulated fluence. The resulting sparsity introduces a reduction in computational time and space proportional to the square of the sparsity tolerance with a potential decrease in cost greater than 10 times that of a dense calculation allowing not only for faster caluclations but for calculations that a dense algorithm could not perform on the same system.

No abstract

In per patient analysis for the first cohort with 50 patients, the percentage of patients who were defined to have oligometastatic state was 70% (35/50) by CAD and 64% (32/50) by the radiologist, respectively (P = 0.377). Respective false-positive and false-negative rate of CAD in determining oligometastatic state compared to the radiologist were 11.4% and 6.7%. The sensitivity of CAD was 96.9% (95% confidence interval (CI), 82.0-99.8) and the specificity of CAD was 77.8% (95% CI, 51.9-92.6). In per lesion analysis among these patients, sensitivity of CAD was 81.6% (95% CI, 74.3-87.2) for nodules equal to or bigger than 3 mm. Respective 5-year survival rates and mean overall survival in the expanded cohort with 305 patients according to the number of nodules were as follows: 75.2%, 106.9 months (95% CI, 89.0-124.8) for patients with a single nodule, 52.9%, 96.1 months (95% CI, 71.5-120.8) for patients with two nodules, 45.7%, 86.9 months (95% CI, 62.4-111.4) for patients with three nodules, 29.1%, 51.2 months (95% CI, 32.4-70.0) for patients with four nodules, and 22.7%, 36.5 months (95% CI, 29.9-43.0) for patients with equal to or more than five nodules. Internal validation of the nomogram based on the data of expanded cohort showed good discrimination with the median time-dependent area under curve at 5-year of 0.830 (Interquartile range, 0.813-0.828). Conclusion: Proper identification of oligo-metastatic state for local ablative therapy with acceptable quantification of metastatic burden was achievable by utilizing CAD.