Potential of a probabilistic framework for target prediction from surrogate respiratory motion during lung radiotherapy

Artificial Intelligence Uncertainty Quantification in Radiotherapy Applications - A Scoping Review

Background/purpose

The use of artificial intelligence (AI) in radiotherapy (RT) is expanding rapidly. However, there exists a notable lack of clinician trust in AI models, underscoring the need for effective uncertainty quantification (UQ) methods. The purpose of this study was to scope existing literature related to UQ in RT, identify areas of improvement, and determine future directions.

Methods

We followed the PRISMA-ScR scoping review reporting guidelines. We utilized the population (human cancer patients), concept (utilization of AI UQ), context (radiotherapy applications) framework to structure our search and screening process. We conducted a systematic search spanning seven databases, supplemented by manual curation, up to January 2024. Our search yielded a total of 8980 articles for initial review. Manuscript screening and data extraction was performed in Covidence. Data extraction categories included general study characteristics, RT characteristics, AI characteristics, and UQ characteristics.

Results

We identified 56 articles published from 2015-2024. 10 domains of RT applications were represented; most studies evaluated auto-contouring (50%), followed by image-synthesis (13%), and multiple applications simultaneously (11%). 12 disease sites were represented, with head and neck cancer being the most common disease site independent of application space (32%). Imaging data was used in 91% of studies, while only 13% incorporated RT dose information. Most studies focused on failure detection as the main application of UQ (60%), with Monte Carlo dropout being the most commonly implemented UQ method (32%) followed by ensembling (16%). 55% of studies did not share code or datasets.

Conclusion

Our review revealed a lack of diversity in UQ for RT applications beyond auto-contouring. Moreover, there was a clear need to study additional UQ methods, such as conformal prediction. Our results may incentivize the development of guidelines for reporting and implementation of UQ in RT.

Online prediction for respiratory movement compensation: a patient-specific gating control for MRI-guided radiotherapy

BACKGROUND

This study aims to validate the effectiveness of linear regression for motion prediction of internal organs or tumors on 2D cine-MR and to present an online gating signal prediction scheme that can improve the accuracy of MR-guided radiotherapy for liver and lung cancer.

MATERIALS AND METHODS

We collected 2D cine-MR sequences of 21 liver cancer patients and 10 lung cancer patients to develop a binary gating signal prediction algorithm that forecasts the crossing-time of tumor motion traces relative to the target threshold. Both 0.4 s and 0.6 s prediction windows were tested using three linear predictors and three recurrent neural networks (RNNs), given the system delay of 0.5 s. Furthermore, an adaptive linear regression model was evaluated using only the first 30 s as the burn-in period, during which the model parameters were adapted during the online prediction process. The accuracy of the predicted traces was measured using amplitude metrics (MAE, RMSE, and R2), and in addition, we proposed three temporal metrics, namely crossing error, gating error, and gating accuracy, which are more relevant to the nature of the gating signals.

RESULTS

In both 0.6 s and 0.4 s prediction cases, linear regression outperformed other methods, demonstrating significantly smaller amplitude errors compared to the RNNs (P < 0.05). The proposed algorithm with adaptive linear regression had the best performance with an average gating accuracy of 98.3% and 98.0%, a gating error of 44 ms and 45 ms, for liver cancer and lung cancer patients, respectively.

CONCLUSION

A functional online gating control scheme was developed with an adaptive linear regression that is both more cost-efficient and accurate than sophisticated RNN based methods in all studied metrics.

Uncertainty-driven determination of target measurement times for indirect tracking validation in adaptive radiotherapy

Objective. Hybrid indirect tumor tracking strategies combine continuous monitoring of surrogate signals with episodic radiographic imaging of the target to check and update their models during the treatment. This validation process is traditionally performed at predetermined and fixed-rate time intervals. This study investigates a new validation procedure based on the real-time uncertainty associated with the predicted target positions.Approach. An adaptive version of a Bayesian method for indirect tracking is developed to simulate different validation processes within a single framework: no validation, regular validation and uncertainty-based validation. While regular validation involves measuring targets at fixed intervals, uncertainty-based validation takes advantage of a key Bayesian feature, which is the real-time confidence information associated with predictions. The validation processes are applied to ground truth breathing signals consisting of a lung target and two different surrogates (one internal, one external). Their impact on prediction accuracy is evaluated with root-mean-square error (RMSE) and incidence of large errors. The number of validation measurements triggered is also examined.Main results. When using the internal surrogate and compared to regular validation, uncertainty-based validation results in significantly better prediction accuracy while using fewer validation measurements: RMSE and fraction of large errors are reduced on average by 12% and 26% respectively, with 36% fewer validation measurements. With the external surrogate, whose correlation with the target is less stable over time, more validation measurements are automatically triggered, which leads to a substantial reduction of prediction errors: RMSE and fraction of large errors are reduced on average by 17% and 28% respectively compared to regular validation. It is also observed that depending on the initial instant, regular validation can result in worse prediction accuracy compared to no validation.Significance. Uncertainty-based validation has the potential to be more efficient and effective than a validation process performed at prescheduled and fixed-rate time intervals.

Predicting respiratory motion using a novel patient specific dual deep recurrent neural networks

Real-time tracking of a target volume is a promising solution for reducing the planning margins and both dosimetric and geometric uncertainties in the treatment of thoracic and upper-abdomen cancers. Respiratory motion prediction is an integral part of real-time tracking to compensate for the latency of tracking systems. The purpose of this work was to develop a novel method for accurate respiratory motion prediction using dual deep recurrent neural networks (RNNs). The respiratory motion data of 111 patients were used to train and evaluate the method. For each patient, two models (Network1 and Network2) were trained on 80% of the respiratory wave, and the remaining 20% was used for evaluation. The first network (Network 1) is a "coarse resolution" prediction of future points and second network (Network 2) provides a "fine resolution" prediction to interpolate between the future predictions. The performance of the method was tested using two types of RNN algorithms : Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU). The accuracy of each model was evaluated using the root mean square error (RMSE) and mean absolute error (MAE). Overall, the RNN model with GRU- function had better accuracy than the RNN model with LSTM-function (RMSE (mm): 0.4±0.2 vs. 0.6±0.3; MAE (mm): 0.4±0.2 vs. 0.6±0.2). The GRU was able to predict the respiratory motion accurately (<1 mm) up to the latency period of 440 ms, and LSTM's accuracy was acceptable only up to 240 ms. The proposed method using GRU function can be used for respiratory-motion prediction up to a latency period of 440 ms.

Prediction of the position of external markers using a recurrent neural network trained with unbiased online recurrent optimization for safe lung cancer radiotherapy

BACKGROUND AND OBJECTIVE

During lung cancer radiotherapy, the position of infrared reflective objects on the chest can be recorded to estimate the tumor location. However, radiotherapy systems have a latency inherent to robot control limitations that impedes the radiation delivery precision. Prediction with online learning of recurrent neural networks (RNN) allows for adaptation to non-stationary respiratory signals, but classical methods such as real-time recurrent learning (RTRL) and truncated backpropagation through time are respectively slow and biased. This study investigates the capabilities of unbiased online recurrent optimization (UORO) to forecast respiratory motion and enhance safety in lung radiotherapy.

METHODS

We used nine observation records of the three-dimensional (3D) position of three external markers on the chest and abdomen of healthy individuals breathing during intervals from 73s to 222s. The sampling frequency was 10Hz, and the amplitudes of the recorded trajectories range from 6mm to 40mm in the superior-inferior direction. We forecast the 3D location of each marker simultaneously with a horizon value (the time interval in advance for which the prediction is made) between 0.1s and 2.0s, using an RNN trained with UORO. We compare its performance with an RNN trained with RTRL, least mean squares (LMS), and offline linear regression. We provide closed-form expressions for quantities involved in the gradient loss calculation in UORO, thereby making its implementation efficient. Training and cross-validation were performed during the first minute of each sequence.

RESULTS

On average over the horizon values considered and the nine sequences, UORO achieves the lowest root-mean-square (RMS) error and maximum error among the compared algorithms. These errors are respectively equal to 1.3mm and 8.8mm, and the prediction time per time step was lower than 2.8ms (Dell Intel core i9-9900K 3.60 GHz). Linear regression has the lowest RMS error for the horizon values 0.1s and 0.2s, followed by LMS for horizon values between 0.3s and 0.5s, and UORO for horizon values greater than 0.6s.

CONCLUSIONS

UORO can accurately predict the 3D position of external markers for intermediate to high response times with an acceptable time performance. This will help limit unwanted damage to healthy tissues caused by radiotherapy.

Offline and online LSTM networks for respiratory motion prediction in MR-guided radiotherapy

Objective. Gated beam delivery is the current clinical practice for respiratory motion compensation in MR-guided radiotherapy, and further research is ongoing to implement tracking. To manage intra-fractional motion using multileaf collimator tracking the total system latency needs to be accounted for in real-time. In this study, long short-term memory (LSTM) networks were optimized for the prediction of superior–inferior tumor centroid positions extracted from clinically acquired 2D cine MRIs. Approach. We used 88 patients treated at the University Hospital of the LMU Munich for training and validation (70 patients, 13.1 h), and for testing (18 patients, 3.0 h). Three patients treated at Fondazione Policlinico Universitario Agostino Gemelli were used as a second testing set (1.5 h). The performance of the LSTMs in terms of root mean square error (RMSE) was compared to baseline linear regression (LR) models for forecasted time spans of 250 ms, 500 ms and 750 ms. Both the LSTM and the LR were trained with offline (offline LSTM and offline LR) and online schemes (offline+online LSTM and online LR), the latter to allow for continuous adaptation to recent respiratory patterns. Main results. We found the offline+online LSTM to perform best for all investigated forecasts. Specifically, when predicting 500 ms ahead it achieved a mean RMSE of 1.20 mm and 1.00 mm, while the best performing LR model achieved a mean RMSE of 1.42 mm and 1.22 mm for the LMU and Gemelli testing set, respectively. Significance. This indicates that LSTM networks have potential as respiratory motion predictors and that continuous online re-optimization can enhance their performance.

Adaptive confidence regions for indirect tracking of moving tumors in radiotherapy

BACKGROUND

Target motion in the course of radiotherapy is one of the largest factors affecting the treatment quality of highly dynamic sites such as lung. A critical component of real-time motion management is not only the prediction of tumor location at a future point in time but assessment of positional uncertainty for the purposes of margin adjustment and optimization of validation schemes.

PURPOSE

In this study, we propose to investigate the ability of a confidence estimator to accurately reflect the reliability of individual target position predictions and prospectively detect large prediction errors by relying exclusively on a surrogate signal.

METHODS

This work uses a Bayesian framework for indirect tracking. While constant covariance estimates are commonly used to express the uncertainty of the models involved, in this study new adaptive estimates are derived from the surrogate behavior to reflect increasing uncertainty when the breathing conditions differ from the reference conditions observed during the training step. The accuracy of the resulting 95% predicted confidence regions (CRs) is evaluated on 9 breathing sequences involving changes of respiratory types (free, thoracic, abdominal, deep). The breathing motions are collected simultaneously from a lung target and two different surrogate signals (an external marker and an anatomical location within the liver). Receiver operating characteristic (ROC) analysis is performed to evaluate the ability of the predicted uncertainty to prospectively detect large prediction errors.

RESULTS

Higher CR accuracy is obtained when using the proposed adaptive estimates over using constant estimations: on average over the cohort, the proportion of actual target positions lying within the 95% CR is increased by 40 and 35 p.p. with the internal and external surrogates. The time-dependent inflation of the CR width tends to match the magnitude variation of the prediction errors : the adaptive CR effectively enlarges when the target position cannot be predicted reliably, which corresponds to potentially high prediction errors. More precisely, the ROC analysis indicates that the proposed uncertainty estimate can detect if prediction errors are greater than 5 mm with on average high sensitivity (90%) and modest specificity (54% and 47% from internal and external surrogates respectively).

CONCLUSIONS

While relying exclusively on the surrogate motion characteristics being continuously monitored, the Bayesian framework coupled to adaptive uncertainty estimations can provide reliable CR able to detect large prediction errors. The findings of this study could be further used to automatically trigger risk management mechanisms prospectively. This article is protected by copyright. All rights reserved.