Comparison of the use of a clinically implemented deep learning segmentation model with the simulated study setting for breast cancer patients receiving radiotherapy.

BACKGROUND: Deep learning (DL) models for auto-segmentation in radiotherapy have been extensively studied in retrospective and pilot settings. However, these studies might not reflect the clinical setting. This study compares the use of a clinically implemented in-house trained DL segmentation model for breast cancer to a previously performed pilot study to assess possible differences in performance or acceptability. MATERIAL AND METHODS: Sixty patients with whole breast radiotherapy, with or without an indication for locoregional radiotherapy were included. Structures were qualitatively scored by radiotherapy technologists and radiation oncologists. Quantitative evaluation was performed using dice-similarity coefficient (DSC), 95th percentile of Hausdorff Distance (95%HD) and surface DSC (sDSC), and time needed for generating, checking, and correcting structures was measured. RESULTS: Ninety-three percent of all contours in clinic were scored as clinically acceptable or usable as a starting point, comparable to 92% achieved in the pilot study. Compared to the pilot study, no significant changes in time reduction were achieved for organs at risks (OARs). For target volumes, significantly more time was needed compared to the pilot study for patients including lymph node levels 1-4, although time reduction was still 33% compared to manual segmentation. Almost all contours have better DSC and 95%HD than inter-observer variations. Only CTVn4 scored worse for both metrics, and the thyroid had a higher 95%HD value. INTERPRETATION: The use of the DL model in clinical practice is comparable to the pilot study, showing high acceptability rates and time reduction.

Breast conserving surgery with intraoperative electron beam radiation therapy for low-risk breast cancer: Five-year follow-up of 306 patients.

Recent studies have reported a higher than expected risk of ipsilateral breast tumor recurrence (IBTR) after breast conserving surgery (BCS) and a single dose of electron beam intra-operative radiotherapy (IORT). This finding was the rationale to perform a retrospective single center cohort study evaluating the oncologic results of consecutive patients treated with BCS and IORT. Women were eligible if they had clinical low-risk (N0, ≤2 cm unifocal, Bloom and Richardson grade 1-2), estrogen receptor-positive and human-epidermal-growth-factor-receptor-2-negative breast cancer. Prior to BCS, pN0 status was determined by sentinel lymph node biopsy. Data on oncologic follow-up were analyzed. Between 2012 and 2019, 306 consecutive patients were treated and analyzed, with a median age of 67 (50-86) years at diagnosis. Median follow-up was 60 (8-120) months. Five-year cumulative risk of IBTR was 13.4% (95% confidence interval [CI] 9.4-17.4). True in field recurrence was present in 3.9% of the patients. In 4.6% of the patients, the IBRT was classified as a local recurrence due to seeding of tumor cells in the cutis or subcutis most likely related to percutaneous biopsy. In 2.9% of the patients, the IBRT was a new outfield primary tumor. Three patients had a regional lymph node recurrence and two had distant metastases as first event. One breast cancer-related death was observed. Estimated 5-year overall survival was 89.8% (95% CI 86.0-93.6). In conclusion, although some of IBTR cases could have been prevented by adaptations in biopsy techniques and patient selection, BCS followed by IORT was associated with a substantial risk of IBTR.

Evaluation of a clinically introduced deep learning model for radiotherapy treatment planning of breast cancer.

Deep learning (DL) models are increasingly studied to automate the process of radiotherapy treatment planning. This study evaluates the clinical use of such a model for whole breast radiotherapy. Treatment plans were automatically generated, after which planners were allowed to manually adapt them. Plans were evaluated based on clinical goals and DVH parameters. Thirty-seven of 50plans did fulfill all clinical goals without adjustments. Thirteen of these 37 plans were still adjusted but did not improve mean heart or lung dose. These results leave room for improvement of both the DL model as well as education on clinically relevant adjustments.

Comparison of the output of a deep learning segmentation model for locoregional breast cancer radiotherapy trained on 2 different datasets.

Introduction: The development of deep learning (DL) models for auto-segmentation is increasing and more models become commercially available. Mostly, commercial models are trained on external data. To study the effect of using a model trained on external data, compared to the same model trained on in-house collected data, the performance of these two DL models was evaluated. Methods: The evaluation was performed using in-house collected data of 30 breast cancer patients. Quantitative analysis was performed using Dice similarity coefficient (DSC), surface DSC (sDSC) and 95th percentile of Hausdorff Distance (95% HD). These values were compared with previously reported inter-observer variations (IOV). Results: For a number of structures, statistically significant differences were found between the two models. For organs at risk, mean values for DSC ranged from 0.63 to 0.98 and 0.71 to 0.96 for the in-house and external model, respectively. For target volumes, mean DSC values of 0.57 to 0.94 and 0.33 to 0.92 were found. The difference of 95% HD values ranged 0.08 to 3.23 mm between the two models, except for CTVn4 with 9.95 mm. For the external model, both DSC and 95% HD are outside the range of IOV for CTVn4, whereas this is the case for the DSC found for the thyroid of the in-house model. Conclusions: Statistically significant differences were found between both models, which were mostly within published inter-observer variations, showing clinical usefulness of both models. Our findings could encourage discussion and revision of existing guidelines, to further decrease inter-observer, but also inter-institute variability.

Clinical evaluation of a deep learning segmentation model including manual adjustments afterwards for locally advanced breast cancer.

Introduction: Deep learning (DL) models are increasingly developed for auto-segmentation in radiotherapy. Qualitative analysis is of great importance for clinical implementation, next to quantitative. This study evaluates a DL segmentation model for left- and right-sided locally advanced breast cancer both quantitatively and qualitatively. Methods: For each side a DL model was trained, including primary breast CTV (CTVp), lymph node levels 1-4, heart, lungs, humeral head, thyroid and esophagus. For evaluation, both automatic segmentation, including correction of contours when needed, and manual delineation was performed and both processes were timed. Quantitative scoring with dice-similarity coefficient (DSC), 95% Hausdorff Distance (95%HD) and surface DSC (sDSC) was used to compare both the automatic (not-corrected) and corrected contours with the manual contours. Qualitative scoring was performed by five radiotherapy technologists and five radiation oncologists using a 3-point Likert scale. Results: Time reduction was achieved using auto-segmentation in 95% of the cases, including correction. The time reduction (mean ± std) was 42.4% ± 26.5% and 58.5% ± 19.1% for OARs and CTVs, respectively, corresponding to an absolute mean reduction (hh:mm:ss) of 00:08:51 and 00:25:38. Good quantitative results were achieved before correction, e.g. mean DSC for the right-sided CTVp was 0.92 ± 0.06, whereas correction statistically significantly improved this contour by only 0.02 ± 0.05, respectively. In 92% of the cases, auto-contours were scored as clinically acceptable, with or without corrections. Conclusions: A DL segmentation model was trained and was shown to be a time-efficient way to generate clinically acceptable contours for locally advanced breast cancer.

Narrowing the difference in dose delivery for IOERT and IOBT for locally advanced and locally recurrent rectal cancer.

Purpose: Intra-operative radiotherapy (IORT) has been used as a tool to provide a high-dose radiation boost to a limited volume of patients with fixed tumors with a likelihood of microscopically involved resection margins, in order to improve local control. Two main techniques to deliver IORT include high-dose-rate (HDR) brachytherapy, termed 'intra-operative brachytherapy' (IOBT), and electrons, termed 'intra-operative electron radiotherapy' (IOERT), both having very different dose distributions. A recent paper described an improved local recurrence-free survival favoring IOBT over IOERT for patients with locally advanced or recurrent rectal cancer and microscopically irradical resections. Although several factors may have contributed to this result, an important difference between the two techniques was the higher surface dose delivered by IOBT. This article described an adaptation of IOERT technique to achieve a comparable surface dose as dose delivered by IOBT. Material and methods: Two steps were taken to increase the surface dose for IOERT: 1. Introducing a bolus to achieve a maximum dose on the surface, and 2. Re-normalizing to allow for the same prescribed dose at reference depth. Conclusions: We describe and propose an adaptation of IOERT technique to increase surface dose, decreasing the differences between these two techniques, with the aim of further improving local control. In addition, an alternative method of dose prescription is suggested, to consider improved comparison with other techniques in the future.

Clinical evaluation of two AI models for automated breast cancer plan generation.

BACKGROUND: Artificial intelligence (AI) shows great potential to streamline the treatment planning process. However, its clinical adoption is slow due to the limited number of clinical evaluation studies and because often, the translation of the predicted dose distribution to a deliverable plan is lacking. This study evaluates two different, deliverable AI plans in terms of their clinical acceptability based on quantitative parameters and qualitative evaluation by four radiation oncologists. METHODS: For 20 left-sided node-negative breast cancer patients, treated with a prescribed dose of 40.05 Gy, using tangential beam intensity modulated radiotherapy, two model-based treatment plans were evaluated against the corresponding manual plan. The two models used were an in-house developed U-net model and a vendor-developed contextual atlas regression forest model (cARF). Radiation oncologists evaluated the clinical acceptability of each blinded plan and ranked plans according to preference. Furthermore, a comparison with the manual plan was made based on dose volume histogram parameters, clinical evaluation criteria and preparation time. RESULTS: The U-net model resulted in a higher average and maximum dose to the PTV (median difference 0.37 Gy and 0.47 Gy respectively) and a slightly higher mean heart dose (MHD) (0.01 Gy). The cARF model led to higher average and maximum doses to the PTV (0.30 and 0.39 Gy respectively) and a slightly higher MHD (0.02 Gy) and mean lung dose (MLD, 0.04 Gy). The maximum MHD/MLD difference was ≤ 0.5 Gy for both AI plans. Regardless of these dose differences, 90-95% of the AI plans were considered clinically acceptable versus 90% of the manual plans. Preferences varied between the radiation oncologists. Plan preparation time was comparable between the U-net model and the manual plan (287 s vs 253 s) while the cARF model took longer (471 s). When only considering user interaction, plan generation time was 121 s for the cARF model and 137 s for the U-net model. CONCLUSIONS: Two AI models were used to generate deliverable plans for breast cancer patients, in a time-efficient manner, requiring minimal user interaction. Although the AI plans resulted in slightly higher doses overall, radiation oncologists considered 90-95% of the AI plans clinically acceptable.

Artificial intelligence based treatment planning of radiotherapy for locally advanced breast cancer.

BACKGROUND AND PURPOSE: Treatment planning of radiotherapy for locally advanced breast cancer patients can be a time consuming process. Artificial intelligence based treatment planning could be used as a tool to speed up this process and maintain plan quality consistency. The purpose of this study was to create treatment plans for locally advanced breast cancer patients using a Convolutional Neural Network (CNN). MATERIALS AND METHODS: Data of 60 patients treated for left-sided breast cancer was used with a training, validation and test split of 36/12/12, respectively. The in-house built CNN model was a hierarchically densely connected U-net (HD U-net). The inputs for the HD U-net were 2D distance maps of the relevant regions of interest. Dose predictions, generated by the HD U-net, were used for a mimicking algorithm in order to create clinically deliverable plans. RESULTS: Dose predictions were generated by the HD U-net and mimicked using a commercial treatment planning system. The predicted plans fulfilling all clinical goals while showing small (≤0.5 Gy) statistically significant differences (p < 0.05) in the doses compared to the manual plans. The mimicked plans show statistically significant differences in the average doses for the heart and lung of ≤0.5 Gy and a reduced D2% of all PTVs. In total, ten of the twelve mimicked plans were clinically acceptable. CONCLUSIONS: We created a CNN model which can generate clinically acceptable plans for left-sided locally advanced breast cancer patients. This model shows great potential to speed up the treatment planning process while maintaining consistent plan quality.

Reduction of heart and lung normal tissue complication probability using automatic beam angle optimization and more generic optimization objectives for breast radiotherapy.

During breast cancer radiotherapy, sparing of healthy tissue is desired. The effect of automatic beam angle optimization and generic dose fall-off objectives on dose and normal tissue complication probabilities was studied. In all patients, dose to lungs and heart showed a mean reduction of 0.4 Gy (range 0.1-1.3 Gy) and 0.2 Gy (range -0.2-0.7 Gy), respectively. These lower doses led to a statistically significant lower cumulative cardiac and lung cancer mortality risk. For smoking patients 40-45 years of age who continue to smoke, it would lead to a reduction from 3.2% ± 0.7% to 2.7% ± 0.6% (p < 0.001).

Development and evaluation of radiotherapy deep learning dose prediction models for breast cancer.

BACKGROUND AND PURPOSE: Treatment planning of radiotherapy is a time-consuming and planner dependent process that can be automated by dose prediction models. The purpose of this study was to evaluate the performance of two machine learning models for breast cancer radiotherapy before possible clinical implementation. MATERIALS AND METHODS: An in-house developed model, based on U-net architecture, and a contextual atlas regression forest (cARF) model integrated in the treatment planning software were trained. Obtained dose distributions were mimicked to create clinically deliverable plans. For training and validation, 90 patients were used, 15 patients were used for testing. Treatment plans were scored on predefined evaluation criteria and percent errors with respect to clinical dose were calculated for doses to planning target volume (PTV) and organs at risk (OARs). RESULTS: The U-net plans before mimicking met all criteria for all patients, both models failed one evaluation criterion in three patients after mimicking. No significant differences (p < 0.05) were found between clinical and predicted U-net plans before mimicking. Doses to OARs in plans of both models differed significantly from clinical plans, but no clinically relevant differences were found. After mimicking, both models had a mean percent error within 1.5% for the average dose to PTV and OARs. The mean errors for maximum doses were higher, within 6.6%. CONCLUSIONS: Differences between predicted doses to OARs of the models were small when compared to clinical plans, and not found to be clinically relevant. Both models show potential in automated treatment planning for breast cancer.

Radiotherapy quality assurance for mesorectum treatment planning within the multi-center phase II STAR-TReC trial: Dutch results.

BACKGROUND: The STAR-TReC trial is an international multi-center, randomized, phase II study assessing the feasibility of short-course radiotherapy or long-course chemoradiotherapy as an alternative to total mesorectal excision surgery. A new target volume is used for both (chemo)radiotherapy arms which includes only the mesorectum. The treatment planning QA revealed substantial variation in dose to organs at risk (OAR) between centers. Therefore, the aim of this study was to determine the treatment plan variability in terms of dose to OAR and assess the effect of a national study group meeting on the quality and variability of treatment plans for mesorectum-only planning for rectal cancer. METHODS: Eight centers produced 25 × 2 Gy treatment plans for five cases. The OAR were the bowel cavity, bladder and femoral heads. A study group meeting for the participating centers was organized to discuss the planning results. At the meeting, the values of the treatment plan DVH parameters were distributed among centers so that results could be compared. Subsequently, the centers were invited to perform replanning if they considered this to be necessary. RESULTS: All treatment plans, both initial planning and replanning, fulfilled the target constraints. Dose to OAR varied considerably for the initial planning, especially for dose levels below 20 Gy, indicating that there was room for trade-offs between the defined OAR. Five centers performed replanning for all cases. One center did not perform replanning at all and two centers performed replanning on two and three cases, respectively. On average, replanning reduced the bowel cavity V20Gy by 12.6%, bowel cavity V10Gy by 22.0%, bladder V35Gy by 14.7% and bladder V10Gy by 10.8%. In 26/30 replanned cases the V10Gy of both the bowel cavity and bladder was lower, indicating an overall lower dose to these OAR instead of a different trade-off. In addition, the bowel cavity V10Gy and V20Gy showed more similarity between centers. CONCLUSIONS: Dose to OAR varied considerably between centers, especially for dose levels below 20 Gy. The study group meeting and the distribution of the initial planning results among centers resulted in lower dose to the defined OAR and reduced variability between centers after replanning. TRIAL REGISTRATION: The STAR-TReC trial, ClinicalTrials.gov Identifier: NCT02945566. Registered 26 October 2016, https://clinicaltrials.gov/ct2/show/NCT02945566).

Interobserver variability in the delineation of the primary lung cancer and lymph nodes on different four-dimensional computed tomography reconstructions.

PURPOSE: The study compared interobserver variation in the delineation of the primary tumour (GTVp) and lymph nodes (GTVln) between three different 4DCT reconstruction types; Maximum Intensity Projection (MIP), Mid-Ventilation (Mid-V) and Mid-Position (Mid-P). MATERIAL AND METHODS: Seven radiation oncologists delineated the GTVp and GTVln on the MIP, Mid-V and Mid-P 4DCT image reconstructions of 10 lung cancer patients. The volumes, the mean standard deviation (SD) and distribution of SD (SD/area) over the median surface contour were compared for different tumour regions. RESULTS: The overall mean delineated volume on the MIP was significantly larger (p < 0.001) than the Mid-V and Mid-P. For the GTVp the Mid-P had the lowest interobserver variation (SD = 0.261 cm), followed by Mid-V (SD = 0.314 cm) and MIP (SD = 0.330 cm) For GTVln the Mid-V had the lowest interobserver variation (SD = 0.425 cm) followed by the MIP (SD = 0.477 cm) and Mid-P (SD = 0.543 cm). The SD/area distribution showed a statistically significant difference between the MIP versus Mid-P and Mid-P versus Mid-V for both GTVp and GTVln (p < 0.001), with outliers indicating interpretation differences for GTVp located close to the mediastinum and GTVln. CONCLUSION: The Mid-P reduced the interobserver variation for the GTVp. Delineation protocols must be improved to benefit from the improved image quality of Mid-P for the GTVln.

Cardiovascular Disease Risk in a Large, Population-Based Cohort of Breast Cancer Survivors.

PURPOSE: To conduct a large, population-based study on cardiovascular disease (CVD) in breast cancer (BC) survivors treated in 1989 or later. METHODS AND MATERIALS: A large, population-based cohort comprising 70,230 surgically treated stage I to III BC patients diagnosed before age 75 years between 1989 and 2005 was linked with population-based registries for CVD. Cardiovascular disease risks were compared with the general population, and within the cohort using competing risk analyses. RESULTS: Compared with the general Dutch population, BC patients had a slightly lower CVD mortality risk (standardized mortality ratio 0.92, 95% confidence interval [CI] 0.88-0.97). Only death due to valvular heart disease was more frequent (standardized mortality ratio 1.28, 95% CI 1.08-1.52). Left-sided radiation therapy after mastectomy increased the risk of any cardiovascular event compared with both surgery alone (subdistribution hazard ratio (sHR) 1.23, 95% CI 1.11-1.36) and right-sided radiation therapy (sHR 1.19, 95% CI 1.04-1.36). Radiation-associated risks were found for not only ischemic heart disease, but also for valvular heart disease and congestive heart failure (CHF). Risks were more pronounced in patients aged <50 years at BC diagnosis (sHR 1.48, 95% CI 1.07-2.04 for left- vs right-sided radiation therapy after mastectomy). Left- versus right-sided radiation therapy after wide local excision did not increase the risk of all CVD combined, yet an increased ischemic heart disease risk was found (sHR 1.14, 95% CI 1.01-1.28). Analyses including detailed radiation therapy information showed an increased CVD risk for left-sided chest wall irradiation alone, left-sided breast irradiation alone, and internal mammary chain field irradiation, all compared with right-sided breast irradiation alone. Compared with patients not treated with chemotherapy, chemotherapy used ≥1997 (ie, anthracyline-based chemotherapy) increased the risk of CHF (sHR 1.35, 95% CI 1.00-1.83). CONCLUSION: Radiation therapy regimens used in BC treatment between 1989 and 2005 increased the risk of CVD, and anthracycline-based chemotherapy regimens increased the risk of CHF.

Dose painting by contours versus dose painting by numbers for stage II/III lung cancer: practical implications of using a broad or sharp brush.

PURPOSE: Local recurrence rates are high in patients with locally advanced NSCLC treated with 60 to 66 Gy in 2 Gy fractions. It is hypothesised that boosting volumes with high SUV on the pre-treatment FDG-PET scan potentially increases local control while maintaining acceptable toxicity levels. We compared two approaches: threshold-based dose painting by contours (DPBC) with voxel-based dose painting by numbers (DPBN). MATERIALS AND METHODS: Two dose painted plans were generated for 10 stage II/III NSCLC patients with 66 Gy at 2-Gy fractions to the entire PTV and a boost dose to the high SUV areas within the primary GTV. DPBC aims for a uniform boost dose at the volume encompassing the SUV 50%-region (GTV(boost)). DPBN aims for a linear relationship between the boost dose to a voxel and the underlying SUV. For both approaches the boost dose was escalated up to 130 Gy (in 33 fractions) or until the dose limiting constraint of an organ at risk was met. RESULTS: For three patients (with relatively small peripheral tumours) the dose within the GTV could be boosted to 130 Gy using both strategies. For the remaining patients the boost dose was confined by a critical structure (mediastinal structures in six patients, lungs in one patient). In general the amount of large brush DPBC boosting is limited whenever the GTV(boost) is close to any serial risk organ. In contrast, small brush DPBN inherently boosts at a voxel-by-voxel basis allowing significant higher dose values to high SUV voxels more distant from the organs at risk. We found that the biological SUV gradients are reasonably congruent with the dose gradients that standard linear accelerators can deliver. CONCLUSIONS: Both large brush DPBC and sharp brush DPBN techniques can be used to considerably boost the dose to the FDG avid regions. However, significantly higher boost levels can be obtained using sharp brush DPBN although sometimes at the cost of a less increased dose to the low SUV regions.

Feasibility of pathology-correlated lung imaging for accurate target definition of lung tumors.

PURPOSE: To accurately define the gross tumor volume (GTV) and clinical target volume (GTV plus microscopic disease spread) for radiotherapy, the pretreatment imaging findings should be correlated with the histopathologic findings. In this pilot study, we investigated the feasibility of pathology-correlated imaging for lung tumors, taking into account lung deformations after surgery. METHODS AND MATERIALS: High-resolution multislice computed tomography (CT) and positron emission tomography (PET) scans were obtained for 5 patients who had non-small-cell lung cancer (NSCLC) before lobectomy. At the pathologic examination, the involved lung lobes were inflated with formalin, sectioned in parallel slices, and photographed, and microscopic sections were obtained. The GTVs were delineated for CT and autocontoured at the 42% PET level, and both were compared with the histopathologic volumes. The CT data were subsequently reformatted in the direction of the macroscopic sections, and the corresponding fiducial points in both images were compared. Hence, the lung deformations were determined to correct the distances of microscopic spread. RESULTS: In 4 of 5 patients, the GTV(CT) was, on average, 4 cm(3) ( approximately 53%) too large. In contrast, for 1 patient (with lymphangitis carcinomatosa), the GTV(CT) was 16 cm(3) ( approximately 40%) too small. The GTV(PET) was too small for the same patient. Regarding deformations, the volume of the well-inflated lung lobes on pathologic examination was still, on average, only 50% of the lobe volume on CT. Consequently, the observed average maximal distance of microscopic spread (5 mm) might, in vivo, be as large as 9 mm. CONCLUSIONS: Our results have shown that pathology-correlated lung imaging is feasible and can be used to improve target definition. Ignoring deformations of the lung might result in underestimation of the microscopic spread.