Dosimetric impact of deep learning-based CT auto-segmentation on radiation therapy treatment planning for prostate cancer

Brand D, Lee E, Loblaw A, Tolan S, van As N, Tree AC. Developing and validating a simple urethra surrogate model to facilitate dosimetric analysis to predict genitourinary toxicity

Purpose

The urethra is a critical structure in prostate radiotherapy planning; however, it is impossible to visualise on CT. We developed a surrogate urethra model (SUM) for CT-only planning workflow and tested its geometric and dosimetric performance against the MRI-delineated urethra (MDU).

Methods

The SUM was compared against 34 different MDUs (within the treatment PTV) in patients treated with 36.25Gy (PTV)/40Gy (CTV) in 5 fractions as part of the PACE-B trial. To assess the surrogate's geometric performance, the Dice similarity coefficient (DSC), Hausdorff distance (HD), mean distance to agreement (MDTA) and the percentage of MDU outside the surrogate (UOS) were calculated. To evaluate the dosimetric performance, a paired t-test was used to calculate the mean of differences between the MDU and SUM for the D99, D98, D50, D2 and D1. The D(n) is the dose (Gy) to n% of the urethra.

Results

The median results showed low agreement on DSC (0.32; IQR 0.21-0.41), but low distance to agreement, as would be expected for a small structure (HD 8.4mm (IQR 7.1-10.1mm), MDTA 2.4mm (IQR, 2.2mm-3.2mm)). The UOS was 30% (IQR, 18-54%), indicating nearly a third of the urethra lay outside of the surrogate. However, when comparing urethral dose between the MDU and SUM, the mean of differences for D99, D98 and D95 were 0.12Gy (p=0.57), 0.09Gy (p=0.61), and 0.11Gy (p=0.46) respectively. The mean of differences between the D50, D2 and D1 were 0.08Gy (p=0.04), 0.09Gy (p=0.02) and 0.1Gy (p=0.01) respectively, indicating good dosimetric agreement between MDU and SUM.

Conclusion

While there were geometric differences between the MDU and SUM, there was no clinically significant difference between urethral dose-volume parameters. This surrogate model could be validated in a larger cohort and then used to estimate the urethral dose on CT planning scans in those without an MRI planning scan or urinary catheter.

Application of deep learning in radiation therapy for cancer

In recent years, with the development of artificial intelligence, deep learning has been gradually applied to clinical treatment and research. It has also found its way into the applications in radiotherapy, a crucial method for cancer treatment. This study summarizes the commonly used and latest deep learning algorithms (including transformer, and diffusion models), introduces the workflow of different radiotherapy, and illustrates the application of different algorithms in different radiotherapy modules, as well as the defects and challenges of deep learning in the field of radiotherapy, so as to provide some help for the development of automatic radiotherapy for cancer.

Patient-Specific Auto-segmentation on Daily kVCT Images for Adaptive Radiation Therapy

PURPOSE

This study explored deep-learning-based patient-specific auto-segmentation using transfer learning on daily RefleXion kilovoltage computed tomography (kVCT) images to facilitate adaptive radiation therapy, based on data from the first group of patients treated with the innovative RefleXion system.

METHODS AND MATERIALS

For head and neck (HaN) and pelvic cancers, a deep convolutional segmentation network was initially trained on a population data set that contained 67 and 56 patient cases, respectively. Then the pretrained population network was adapted to the specific RefleXion patient by fine-tuning the network weights with a transfer learning method. For each of the 6 collected RefleXion HaN cases and 4 pelvic cases, initial planning computed tomography (CT) scans and 5 to 26 sets of daily kVCT images were used for the patient-specific learning and evaluation separately. The performance of the patient-specific network was compared with the population network and the clinical rigid registration method and evaluated by the Dice similarity coefficient (DSC) with manual contours being the reference. The corresponding dosimetric effects resulting from different auto-segmentation and registration methods were also investigated.

RESULTS

The proposed patient-specific network achieved mean DSC results of 0.88 for 3 HaN organs at risk (OARs) of interest and 0.90 for 8 pelvic target and OARs, outperforming the population network (0.70 and 0.63) and the registration method (0.72 and 0.72). The DSC of the patient-specific network gradually increased with the increment of longitudinal training cases and approached saturation with more than 6 training cases. Compared with using the registration contour, the target and OAR mean doses and dose-volume histograms obtained using the patient-specific auto-segmentation were closer to the results using the manual contour.

CONCLUSIONS

Auto-segmentation of RefleXion kVCT images based on the patient-specific transfer learning could achieve higher accuracy, outperforming a common population network and clinical registration-based method. This approach shows promise in improving dose evaluation accuracy in RefleXion adaptive radiation therapy.

Clinical evaluation of deep learning-based automatic clinical target volume segmentation: a single-institution multi-site tumor experience

PURPOSE

The large variability in tumor appearance and shape makes manual delineation of the clinical target volume (CTV) time-consuming, and the results depend on the oncologists' experience. Whereas deep learning techniques have allowed oncologists to automate the CTV delineation, multi-site tumor analysis is often lacking in the literature. This study aimed to evaluate the deep learning models that automatically contour CTVs of tumors at various sites on computed tomography (CT) images from objective and subjective perspectives.

METHODS AND MATERIALS

577 patients were selected for the present study, including nasopharyngeal (n = 34), esophageal (n = 40), breast-conserving surgery (BCS) (left-sided, n = 71; right-sided, n = 71), breast-radical mastectomy (BRM) (left-sided, n = 43; right-sided, n = 37), cervical (radical radiotherapy, n = 45; postoperative, n = 85), prostate (n = 42), and rectal (n = 109) carcinomas. Manually delineated CTV contours by radiation oncologists are served as ground truth. Four models were evaluated: Flexnet, Unet, Vnet, and Segresnet, which are commercially available in the medical product "AccuLearning AI model training platform". The data were divided into the training, validation, and testing set at a ratio of 5:1:4. The geometric metrics, including Dice Similarity Coefficient (DSC) and Hausdorff Distance (HD), were calculated for objective evaluation. For subjective assessment, oncologists rated the segmentation contours of the testing set visually.

RESULTS

High correlations were observed between automatic and manual contours. Based on the results of the independent test group, most of the patients achieved satisfactory quantitative results (DSC > 0.8), except for patients with esophageal carcinoma (DSC: 0.62-0.64). The subjective review indicated that 82.65% of predicted CTVs scored either as clinically accepting (8.68%) or requiring minor revision (73.97%), and no patients were scored as rejected.

CONCLUSION

This experimental work demonstrated that auto-generated contours could serve as an initial template to help oncologists save time in CTV delineation. The deep learning-based auto-segmentations achieve acceptable accuracy and show the potential to improve clinical efficiency for radiotherapy of a variety of cancer.

Deep learning based automatic segmentation of organs-at-risk for 0.35 T MRgRT of lung tumors

BACKGROUND AND PURPOSE

Magnetic resonance imaging guided radiotherapy (MRgRT) offers treatment plan adaptation to the anatomy of the day. In the current MRgRT workflow, this requires the time consuming and repetitive task of manual delineation of organs-at-risk (OARs), which is also prone to inter- and intra-observer variability. Therefore, deep learning autosegmentation (DLAS) is becoming increasingly attractive. No investigation of its application to OARs in thoracic magnetic resonance images (MRIs) from MRgRT has been done so far. This study aimed to fill this gap.

MATERIALS AND METHODS

122 planning MRIs from patients treated at a 0.35 T MR-Linac were retrospectively collected. Using an 80/19/23 (training/validation/test) split, individual 3D U-Nets for segmentation of the left lung, right lung, heart, aorta, spinal canal and esophagus were trained. These were compared to the clinically used contours based on Dice similarity coefficient (DSC) and Hausdorff distance (HD). They were also graded on their clinical usability by a radiation oncologist.

RESULTS

Median DSC was 0.96, 0.96, 0.94, 0.90, 0.88 and 0.78 for left lung, right lung, heart, aorta, spinal canal and esophagus, respectively. Median 95th percentile values of the HD were 3.9, 5.3, 5.8, 3.0, 2.6 and 3.5 mm, respectively. The physician preferred the network generated contours over the clinical contours, deeming 85 out of 129 to not require any correction, 25 immediately usable for treatment planning, 15 requiring minor and 4 requiring major corrections.

CONCLUSIONS

We trained 3D U-Nets on clinical MRI planning data which produced accurate delineations in the thoracic region. DLAS contours were preferred over the clinical contours.

Deep-learning magnetic resonance imaging-based automatic segmentation for organs-at-risk in the brain: Accuracy and impact on dose distribution

Background and purpose

Normal tissue sparing in radiotherapy relies on proper delineation. While manual contouring is time consuming and subject to inter-observer variability, auto-contouring could optimize workflows and harmonize practice. We assessed the accuracy of a commercial, deep-learning, MRI-based tool for brain organs-at-risk delineation.

Materials and methods

Thirty adult brain tumor patients were retrospectively manually recontoured. Two additional structure sets were obtained: AI (artificial intelligence) and AIedit (manually corrected auto-contours). For 15 selected cases, identical plans were optimized for each structure set. We used Dice Similarity Coefficient (DSC) and mean surface-distance (MSD) for geometric comparison and gamma analysis and dose-volume-histogram comparison for dose metrics evaluation. Wilcoxon signed-ranks test was used for paired data, Spearman coefficient(ρ) for correlations and Bland-Altman plots to assess level of agreement.

Results

Auto-contouring was significantly faster than manual (1.1/20 min, p < 0.01). Median DSC and MSD were 0.7/0.9 mm for AI and 0.8/0.5 mm for AIedit. DSC was significantly correlated with structure size (ρ = 0.76, p < 0.01), with higher DSC for large structures. Median gamma pass rate was 74% (71-81%) for Plan_AI and 82% (75-86%) for Plan_AIedit, with no correlation with DSC or MSD. Differences between Dmean_AI and Dmean_Ref were ≤ 0.2 Gy (p < 0.05). The dose difference was moderately correlated with DSC. Bland Altman plot showed minimal discrepancy (0.1/0) between AI and reference Dmean/Dmax.

Conclusions

The AI-model showed good accuracy for large structures, but developments are required for smaller ones. Auto-segmentation was significantly faster, with minor differences in dose distribution caused by geometric variations.

Evaluation of therapeutic radiographer contouring for magnetic resonance image guided online adaptive prostate radiotherapy

BACKGROUND AND PURPOSE

The implementation of MRI-guided online adaptive radiotherapy has facilitated the extension of therapeutic radiographers' roles to include contouring, thus releasing the clinician from attending daily treatment. Following undergoing a specifically designed training programme, an online interobserver variability study was performed.

MATERIALS AND METHODS

117 images from six patients treated on a MR Linac were contoured online by either radiographer or clinician and the same images contoured offline by the alternate profession. Dice similarity coefficient (DSC), mean distance to agreement (MDA), Hausdorff distance (HD) and volume metrics were used to analyse contours. Additionally, the online radiographer contours and optimised plans (n = 59) were analysed using the offline clinician defined contours. After clinical implementation of radiographer contouring, target volume comparison and dose analysis was performed on 20 contours from five patients.

RESULTS

Comparison of the radiographers' and clinicians' contours resulted in a median (range) DSC of 0.92 (0.86 - 0.99), median (range) MDA of 0.98 mm (0.2-1.7) and median (range) HD of 6.3 mm (2.5-11.5) for all 117 fractions. There was no significant difference in volume size between the two groups. Of the 59 plans created with radiographer online contours and overlaid with clinicians' offline contours, 39 met mandatory dose constraints and 12 were acceptable because 95 % of the high dose PTV was covered by 95 % dose, or the high dose PTV was within 3 % of online plan. A clinician blindly reviewed the eight remaining fractions and, using trial quality assurance metrics, deemed all to be acceptable. Following clinical implementation of radiographer contouring, the median (range) DSC of CTV was 0.93 (0.88-1.0), median (range) MDA was 0.8 mm (0.04-1.18) and HD was 5.15 mm (2.09-8.54) respectively. Of the 20 plans created using radiographer online contours overlaid with clinicians' offline contours, 18 met the dosimetric success criteria, the remaining 2 were deemed acceptable by a clinician.

CONCLUSION

Radiographer and clinician prostate and seminal vesicle contours on MRI for an online adaptive workflow are comparable and produce clinically acceptable plans. Radiographer contouring for prostate treatment on a MR-linac can be effectively introduced with appropriate training and evaluation. A DSC threshold for target structures could be implemented to streamline future training.

Interobserver variation of clinical oncologists compared to therapeutic radiographers (RTT) prostate contours on T2 weighted MRI

The implementation of MRI-guided online adaptive radiotherapy has enabled extension of therapeutic radiographers' roles to include contouring. An offline interobserver variability study compared five radiographers' and five clinicians' contours on 10 MRIs acquired on a MR-Linac from 10 patients. All contours were compared to a "gold standard" created from an average of clinicians' contours. The median (range) DSC of radiographers' and clinicians' contours compared to the "gold standard" was 0.91 (0.86-0.96), and 0.93 (0.88-0.97) respectively illustrating non-inferiority of the radiographers' contours to the clinicians. There was no significant difference in HD, MDA or volume size between the groups.

Evaluation of real-time tumor contour prediction using LSTM networks for MR-guided radiotherapy

BACKGROUND AND PURPOSE

Magnetic resonance imaging guided radiotherapy (MRgRT) with deformable multileaf collimator (MLC) tracking would allow to tackle both rigid displacement and tumor deformation without prolonging treatment. However, the system latency must be accounted for by predicting future tumor contours in real-time. We compared the performance of three artificial intelligence (AI) algorithms based on long short-term memory (LSTM) modules for the prediction of 2D-contours 500ms into the future.

MATERIALS AND METHODS

Models were trained (52 patients, 3.1h of motion), validated (18 patients, 0.6h) and tested (18 patients, 1.1h) with cine MRs from patients treated at one institution. Additionally, we used three patients (2.9h) treated at another institution as second testing set. We implemented 1) a classical LSTM network (LSTM-shift) predicting tumor centroid positions in superior-inferior and anterior-posterior direction which are used to shift the last observed tumor contour. The LSTM-shift model was optimized both in an offline and online fashion. We also implemented 2) a convolutional LSTM model (ConvLSTM) to directly predict future tumor contours and 3) a convolutional LSTM combined with spatial transformer layers (ConvLSTM-STL) to predict displacement fields used to warp the last tumor contour.

RESULTS

The online LSTM-shift model was found to perform slightly better than the offline LSTM-shift and significantly better than the ConvLSTM and ConvLSTM-STL. It achieved a 50% Hausdorff distance of 1.2mm and 1.0mm for the two testing sets, respectively. Larger motion ranges were found to lead to more substantial performance differences across the models.

CONCLUSION

LSTM networks predicting future centroids and shifting the last tumor contour are the most suitable for tumor contour prediction. The obtained accuracy would allow to reduce residual tracking errors during MRgRT with deformable MLC-tracking.

Automated Contouring and Planning in Radiation Therapy: What Is 'Clinically Acceptable'?

Developers and users of artificial-intelligence-based tools for automatic contouring and treatment planning in radiotherapy are expected to assess clinical acceptability of these tools. However, what is 'clinical acceptability'? Quantitative and qualitative approaches have been used to assess this ill-defined concept, all of which have advantages and disadvantages or limitations. The approach chosen may depend on the goal of the study as well as on available resources. In this paper, we discuss various aspects of 'clinical acceptability' and how they can move us toward a standard for defining clinical acceptability of new autocontouring and planning tools.