Auto-segmentation Method Research Articles

A multicenter study on deep learning for glioblastoma auto-segmentation with prior knowledge in multimodal imaging.

A precise radiotherapy plan is crucial to ensure accurate segmentation of glioblastomas (GBMs) for radiation therapy. However, the traditional manual segmentation process is labor-intensive and heavily reliant on the experience of radiation oncologists. In this retrospective study, a novel auto-segmentation method is proposed to address these problems. To assess the method's applicability across diverse scenarios, we conducted its development and evaluation using a cohort of 148 eligible patients drawn from four multicenter datasets and retrospective data collection including noncontrast CT, multisequence MRI scans, and corresponding medical records. All patients were diagnosed with histologically confirmed high-grade glioma (HGG). A deep learning-based method (PKMI-Net) for automatically segmenting gross tumor volume (GTV) and clinical target volumes (CTV1 and CTV2) of GBMs was proposed by leveraging prior knowledge from multimodal imaging. The proposed PKMI-Net demonstrated high accuracy in segmenting, respectively, GTV, CTV1, and CTV2 in an 11-patient test set, achieving Dice similarity coefficients (DSC) of 0.94, 0.95, and 0.92; 95% Hausdorff distances (HD95) of 2.07, 1.18, and 3.95 mm; average surface distances (ASD) of 0.69, 0.39, and 1.17 mm; and relative volume differences (RVD) of 5.50%, 9.68%, and 3.97%. Moreover, the vast majority of GTV, CTV1, and CTV2 produced by PKMI-Net are clinically acceptable and require no revision for clinical practice. In our multicenter evaluation, the PKMI-Net exhibited consistent and robust generalizability across the various datasets, demonstrating its effectiveness in automatically segmenting GBMs. The proposed method using prior knowledge in multimodal imaging can improve the contouring accuracy of GBMs, which holds the potential to improve the quality and efficiency of GBMs' radiotherapy.

Refining ICESAT-2 ATL13 Altimetry Data for Improving Water Surface Elevation Accuracy on Rivers

The application of ICESAT-2 altimetry data in river hydrology critically depends on the accuracy of the mean water surface elevation (WSE) at a virtual station (VS) where satellite observations intersect solely with water. It is acknowledged that the ATL13 product has noise elevations of the adjacent land, resulting in biased high mean WSEs at VSs. Earlier studies have relied on human intervention or water masks to resolve this. Both approaches are unsatisfactory solutions for large river basins where the issue becomes pronounced due to many tributaries and meanders. There is no automated procedure to partition the truly representative water height from the totality of the along-track ICESAT-2 photon segments (portions of photon points along a beam) for increasing precision of the mean WSE at VSs. We have developed an automated approach called “auto-segmentation”. The accuracy of our method was assessed by comparing the ATL13-derived WSEs with direct water level observations at 10 different gauging stations on 37 different dates along the Lower Murray River, Australia. The concordance between the two datasets is significantly high and without detectable bias. In addition, we evaluated the effects of four methods for calculating the mean WSEs at VSs after auto-segmentation processing. Our results reveal that all methods perform almost equally well, with the same R2 value (0.998) and only subtle variations in RMSE (0.181–0.189 m) and MAE (0.130–0.142 m). We also found that the R2, RMSE and MAE are better under the high flow condition (0.999, 0.124 and 0.111 m) than those under the normal-low flow condition (0.997, 0.208 and 0.160 m). Overall, our auto-segmentation method is an effective and efficient approach for deriving accurate mean WSEs at river VSs. It will contribute to the improvement of ICESAT-2 ATL13 altimetry data utility on rivers.

Semi-supervised auto-segmentation method for pelvic organ-at-risk in magnetic resonance images based on deep-learning.

In radiotherapy, magnetic resonance (MR) imaging has higher contrast for soft tissues compared to computed tomography (CT) scanning and does not emit radiation. However, manual annotation of the deep learning-based automatic organ-at-risk (OAR) delineation algorithms is expensive, making the collection of large-high-quality annotated datasets a challenge. Therefore, we proposed the low-cost semi-supervised OAR segmentation method using small pelvic MR image annotations. We trained a deep learning-based segmentation model using 116 sets of MR images from 116 patients. The bladder, femoral heads, rectum, and small intestine were selected as OAR regions. To generate the training set, we utilized a semi-supervised method and ensemble learning techniques. Additionally, we employed a post-processing algorithm to correct the self-annotation data. Both 2D and 3D auto-segmentation networks were evaluated for their performance. Furthermore, we evaluated the performance of semi-supervised method for 50 labeled data and only 10 labeled data. The Dice similarity coefficient (DSC) of the bladder, femoral heads, rectum and small intestine between segmentation results and reference masks is 0.954, 0.984, 0.908, 0.852 only using self-annotation and post-processing methods of 2D segmentation model. The DSC of corresponding OARs is 0.871, 0.975, 0.975, 0.783, 0.724 using 3D segmentation network, 0.896, 0.984, 0.890, 0.828 using 2D segmentation network and common supervised method. The outcomes of our study demonstrate that it is possible to train a multi-OAR segmentation model using small annotation samples and additional unlabeled data. To effectively annotate the dataset, ensemble learning and post-processing methods were employed. Additionally, when dealing with anisotropy and limited sample sizes, the 2D model outperformed the 3D model in terms of performance.

Development of statistical auto-segmentation method for diffusion restriction gray matter lesions in patients with newly diagnosed sporadic Creutzfeldt–Jakob disease

Quantification of diffusion restriction lesions in sporadic Creutzfeldt-Jakob disease (sCJD) may provide information of the disease burden. We aim to develop an automatic segmentation model for sCJD and to evaluate the volume of disease extent as a prognostic marker for overall survival. Fifty-six patients (mean age ± SD, 61.2 ± 9.9 years) were included from February 2000 to July 2020. A threshold-based segmentation was used to obtain abnormal signal intensity masks. Segmented volumes were compared with the visual grade. The Dice similarity coefficient was calculated to measure the similarity between the automatic vs. manual segmentation. Cox proportional hazards regression analysis was performed to evaluate the volume of disease extent as a prognostic marker. The automatic segmentation showed good correlation with the visual grading. The cortical lesion volumes significantly increased as the visual grade aggravated (extensive: 112.9 ± 73.2; moderate: 45.4 ± 30.4; minimal involvement: 29.6 ± 18.1 mm3) (P < 0.001). The deep gray matter lesion volumes were significantly higher for positive than for negative involvement of the deep gray matter (5.6 ± 4.6 mm3 vs. 1.0 ± 1.3 mm3, P < 0.001). The mean Dice similarity coefficients were 0.90 and 0.94 for cortical and deep gray matter lesions, respectively. However, the volume of disease extent was not associated with worse overall survival (cortical extent: P = 0.07; deep gray matter extent: P = 0.12).

Gaussian filter facilitated deep learning-based architecture for accurate and efficient liver tumor segmentation for radiation therapy.

Addressing the challenges of unclear tumor boundaries and the confusion between cysts and tumors in liver tumor segmentation, this study aims to develop an auto-segmentation method utilizing Gaussian filter with the nnUNet architecture to effectively distinguish between tumors and cysts, enhancing the accuracy of liver tumor auto-segmentation. Firstly, 130 cases of liver tumorsegmentation challenge 2017 (LiTS2017) were used for training and validating nnU-Net-based auto-segmentation model. Then, 14 cases of 3D-IRCADb dataset and 25 liver cancer cases retrospectively collected in our hospital were used for testing. The dice similarity coefficient (DSC) was used to evaluate the accuracy of auto-segmentation model by comparing with manual contours. The nnU-Net achieved an average DSC value of 0.86 for validation set (20 LiTS cases) and 0.82 for public testing set (14 3D-IRCADb cases). For clinical testing set, the standalone nnU-Net model achieved an average DSC value of 0.75, which increased to 0.81 after post-processing with the Gaussian filter (P<0.05), demonstrating its effectiveness in mitigating the influence of liver cysts on liver tumor segmentation. Experiments show that Gaussian filter is beneficial to improve the accuracy of liver tumor segmentation in clinic.

An open source auto-segmentation algorithm for delineating heart and substructures – Development and validation within a multicenter lung cancer cohort

Background and purposeIrradiation of the heart in thoracic cancers raises toxicity concerns. For accurate dose estimation, automated heart and substructure segmentation is potentially useful. In this study, a hybrid automatic segmentation is developed. The accuracy of delineation and dose predictions were evaluated, testing the method's potential within heart toxicity studies. Materials and MethodsThe hybrid segmentation method delineated the heart, four chambers, three large vessels, and the coronary arteries. The method consisted of a nnU-net heart segmentation and partly atlas- and model-based segmentation of the substructures. The nnU-net training and atlas segmentation was based on lung cancer patients and was validated against a national consensus dataset of 12 patients with breast cancer. The accuracy of dose predictions between manual and auto-segmented heart and substructures was evaluated by transferring the dose distribution of 240 previously treated lung cancer patients to the consensus data set. ResultsThe hybrid auto-segmentation method performed well with a heart dice similarity coefficient (DSC) of 0.95, with no statistically significant difference between the automatic and manual delineations. The DSC for the chambers varied from 0.78-0.86 for the automatic segmentation and was comparable with the inter-observer variability. Most importantly, the automatic segmentation was as precise as the clinical experts in predicting the dose distribution to the heart and all substructures. ConclusionThe hybrid segmentation method performed well in delineating the heart and substructures. The prediction of dose by the automatic segmentation was aligned with the manual delineations, enabling measurement of heart and substructure dose in large cohorts. The delineation algorithm will be available for download.

Machine Learning Approach for Diagnosis and Prognosis of Cardiac Arrhythmia Condition Using a Minimum Feature Set and Auto-Segmentation-Based Window Optimisation

Cardiovascular diseases have become extremely prevalent in the global population. Several accurate classification methods for arrhythmias have been proposed in the healthcare literature. However, extensive research is required to improve the prediction accuracy of various arrhythmia conditions. In this paper, discussion is focussed on two major objectives: optimisation of windows based on our proposed auto-segmentation method for the exact diagnosis of the heart condition within the segment and prediction of arrhythmia progression. For prediction, identification of features is vital. Identified efficient independent feature sets such as RR interval, peak-to-peak amplitude, and unique derived parameters such as coefficient of variation (CV) of RR interval and CV of peak-to-peak amplitude. The progression of arrhythmia includes the following steps such as data preprocessing, time and frequency domain feature extraction, and feature selection using principal component analysis. A hypertuned support vector machine is utilised for accurate diagnosis. Proposed two techniques to predict the progression of arrhythmias: the regression-based trend curve (RBTC) and the fuzzy enhanced Markov model (FEMM). We have effectively evaluated our prediction algorithms using offline Massachusetts Institute of Technology Physio Net database signals, using automatic segmentation with prediction accuracy of 98 %. In terms of accuracy, FEMM outperforms RBTC. Thus, an auto-segmentation algorithm was proposed to classify various arrhythmia signals using a minimal feature set and to predict future conditions using our proposed method, FEMM.

Deep-Learning Based Auto-Segmentation for Liver Yttrium-90 Selective Internal Radiation Therapy

In resin Yttrium-90 (Y-90) selective internal radiation therapy (SIRT), liver volume sizes are needed for Y-90 activity calculations using the body-surface-area method, which are obtained from contours that are manually delineated in 3D images. The aim was to apply a deep-learning based auto-segmentation method for liver delineation for Y-90 SIRT. A deep-learning-based liver segmentation method was applied using the U-Net3D architecture, which is a 3D convolutional neural network (CNN) extended from the original 2D U-Net architecture for 3D objects in medical imaging. The segmentation model was trained on the Liver Tumor Segmentation (LiTS) dataset. The training data set contained 130 CT scans, and the test data set contained 70 CT scans. The model was deployed in the clinic using DICOM communication. Auto-segmentation of liver in the CT images of 18 SIRT patients was studied. The CT images were exported from clinical database to the segmentation model's DICOM location, where a monitoring software detected the incoming data and automatically ran the liver segmentation. The results were then returned to the original DICOM location where the CT images were stored. Auto-segmented liver contours were compared with physician manually-delineated contours. Dice similarity coefficient (DSC), mean distance to agreement (MDA), ratio of volume (RV), and ratio of activity (RA, ratio of activity calculated using an auto-segmented liver contour to the accurate activity calculated using a manually-delineated contour), were assessed. DSC, MDA, and RV are 0.942±0.014 (range: 0.908-0.959), 1.902±0.503 mm (range: 1.043-2.956 mm), and 0.988±0.039 (range: 0.901-1.045), respectively. RA is 1.001±0.003 (range: 0.993-1.007), which indicates that the activities calculated using the auto-segmented liver contours are close to the accurate activities. The segmentation model was able to successfully identify and segment livers in the CT images, and provide accurate and reliable results. The proposed method is beneficial for clinical use as it can process large amounts of data quickly and efficiently, and can be easily deployed in a clinical environment using DICOM communication.

Small Samples and Low-Cost Auto-Segmentation Method for Pelvic Organ-at-Risk Segmentation in Magnetic Resonance Images Using Deep-Learning

In radiotherapy, magnetic resonance (MR) imaging has higher contrast of soft tissue, and no radiation compared with computed tomography (CT) scanning. Due to the high-cost of manual annotation, the deep-learning based automatic organ-at-risk (OAR) and target delineation algorithms are in high-demand, but the collecting of large amounts of high-quality annotated datasets remains difficulty. In this paper, we proposed a low-cost OAR segmentation method with semi-supervised annotation using small annotation samples of pelvic MR images. This study consisted of 94 patients diagnosed with rectal cancer from April 2018 to March 2021 at Peking University People's Hospital. We used 17 slices of MR images with annotation and 78 slices without annotation to train a deep-learning based segmentation model. The bladder, femoral heads, rectum and small intestine were selected as OAR. Semi-supervised method and ensemble learning were used for generating training set using small sample with annotation. Post-processing algorithm was used to correct the self-annotation data. Two of 14 annotation samples were set as test set. As for un-labeled images, 40 of them were set as semi-supervised annotation train set, the rest were test set. Besides, both 2D and 3D auto-segmentation networks were evaluated. The dice of bladder, femoral head left and right, rectum and small intestine between segmentation results and reference masks is 0.947, 0.983, 0.981, 0.900, 0.845 only using self-annotation and post-processing method of 2D segmentation model. And the dice of corresponding OAR is 0.871, 0.975, 0.975, 0.783, 0.724 using 3D segmentation network, 0.885,0.982, 0.982, 0.882, 0,814 using 2D segmentation network with supervised method (nnUNet). The 2D model outperformed 3D model with better segmentation performance, shorter inference time and fewer parameters. The results proved that we can train a multi-OAR segmentation model only using small annotation samples and other unlabeled samples. Ensemble learning and post-processing methods are necessary for semi-supervised data annotation. For anisotropy data, 2D model shows better performance than 3D models.

Adaptive Auto-Segmentation for MRI-Guided Online Adaptive Radiotherapy of Cervical Carcinoma

Accurate and efficient delineation of organs and targets on session images is critical in MRI-guided online adaptive radiotherapy (MRgOART). This study proposes a registration-guided deep learning image segmentation framework to assist online delineation of cervical carcinoma. A total of 300 T2-weighted MR images were acquired for patients with cervical carcinoma treated by a 1.5T Unity MR-Linac. The CTV, bladder, rectum, pelvic bone and femoral joints were delineated on each MRI by the same radiation oncologist. To overcome these obstacles to online MRI segmentation, we propose a registration-guided DL (RgDL) segmentation framework that integrates image registration algorithms and DL segmentation models. Firstly, the DL segmentation model was trained using nnU-net. Then, for each treatment fraction, the deformable image registration (DIR) algorithm generates initial contours from previous treatment fraction, which were used as guidance by DL model to obtain the accurate current segmentation. The segmentation accuracy of alone DIR, DL and RgDL were evaluated by dice similarity coefficients (DSC) and other distance-based metrics. Compared to the baseline approaches using the DIR and the DL alone, RgDL achieved a DSC of 91.12% on CTV, higher than DIR and DL alone by 15.54% and 10.13%. The DSC of RgDL were improved to 95.58%, 93.65%, 87.8% and 94.84% for bladder, pelvic bone, rectum and femoral joints, higher than DIR and DL alone by 9.61% on average. The proposed adaptive auto-segmentation method can achieve accurate and efficient segmentation and potentially overcome these obstacles to MRgOART.

An efficient strategy to select head and neck cancer patients for adaptive radiotherapy

Background and purposeAdaptive radiotherapy (ART) is workload intensive but only benefits a subgroup of patients. We aimed to develop an efficient strategy to select candidates for ART in the first two weeks of head and neck cancer (HNC) radiotherapy. Materials and methodsThis study retrospectively enrolled 110 HNC patients who underwent modern photon radiotherapy with at least 5 weekly in-treatment re-scan CTs. A semi auto-segmentation method was applied to obtain the weekly mean dose (Dmean) to OARs. A comprehensive NTCP-profile was applied to obtain NTCP’s. The difference between planning and actual values of Dmean (ΔDmean) and dichotomized difference of clinical relevance (BIOΔNTCP) were used for modelling to determine the cut-off maximum ΔDmean of OARs in week 1 and 2 (maxΔDmean_1 and maxΔDmean_2). Four strategies to select candidates for ART, using cut-off maxΔDmean were compared. ResultsThe Spearman’s rank correlation test showed significant positive correlation between maxΔDmean and BIOΔNTCP (p-value <0.001). For major BIOΔNTCP (>5%) of acute and late toxicity, 10.9% and 4.5% of the patients were true candidates for ART. Strategy C using both cut-off maxΔDmean_1 (3.01 and 5.14 Gy) and cut-off maxΔDmean_2 (3.41 and 5.30 Gy) showed the best sensitivity, specificity, positive and negative predictive values (0.92, 0.82, 0.38, 0.99 for acute toxicity and 1.00, 0.92, 0.38, 1.00 for late toxicity, respectively). ConclusionsWe propose an efficient selection strategy for ART that is able to classify the subgroup of patients with >5% BIOΔNTCP for late toxicity using imaging in the first two treatment weeks.

Patient-Specific Daily Updated Deep Learning Auto-Segmentation for MRI-Guided Adaptive Radiotherapy

<h3>Purpose/Objective(s)</h3> MR-guided adaptive radiotherapy (MRgART) requires accurate and efficient daily image segmentation for plan adaption. Deep learning (DL) technique has shown great potential on medical image segmentation, but still has limited success on MRIs, partially due to the large variations for MRIs from different scanners, protocols, and patients. The longitudinal MRIs and contours during MRgART contain the anatomy variation for the specific patient. However, it has been ignored in current auto-segmentation solutions. This study proposed a patient-specific DL auto-segmentation strategy using patient's previous images and contours to update a prototype model, to improve segmentation accuracy and efficiency for MRgART. <h3>Materials/Methods</h3> For each patient, a prototype model was trained based on SegResNet using the first set of MRI scan and the corresponding contours as inputs. The patient-specific prototype model was updated with each newly added fractional MRI/contours and then used to predict contours for next MRI of the specific patient. During model training, a mutant of the prototype tried to fit the patient's previous data under consistency constraints, which is to limit the difference between the predictions for the latest MRI (volume, length, centroid) within a reasonable range. Prototype weights were partially inherited from the mutant using stochastic weight averaging to get the final model. Six abdominal or pelvic patients (each with 8 longitudinal MRIs) underwent MRgART on a 1.5T MRI-Linac were utilized. A total of 16 abdominal and pelvic organ-at-risks (OARs) and the tumors were manually contoured by three senior radiation oncologists and used as ground truth. The model performance was compared with deformable image registration (DIR) and frozen DL model using Dice Similarity Coefficient (DSC). The contouring time including model inference and contour reviewing and editing time until physician's approval was recorded and compared with fully manual contouring. <h3>Results</h3> The proposed daily updated model achieved superior performance with average DSC of 0.907±0.117, as compared to DIR (0.874±0.109) and frozen DL models (0.889±0.121). As for tumors, the proposed method achieved an average DSC of 0.922±0.097, which is significantly higher than those from DIR (0.723±0.183) and frozen DL (0.591±0.232). The DSC increased by average 3.1% with one additional prior images/contours compared with the prototype model. The contouring time is significant shorter using proposed method (average 6 minutes) than the manual process (20∼40 minutes). <h3>Conclusion</h3> The proposed patient-specific DL auto segmentation method can significantly improve the efficiency and accuracy of contour generation on longitudinal MRIs. It can be integrated into the daily adaption workflow, accelerating the contouring process, thereby facilitating routine practice of MRgART.

Patient-Specific Auto-Segmentation of Target and OARs via Deep Learning on Daily Fan-Beam CT for Adaptive Prostate Radiotherapy

<h3>Purpose/Objective(s)</h3> Contouring the target and organs at risk (OARs) of the treated patient in each fraction is a key step of daily adaptive radiotherapy. We designed a patient-specific deep learning study to improve the auto-segmentation performance for adaptive prostate radiotherapy, by leveraging the daily image-guidance fan-beam CT images of the specific patient available in the current IGRT workflow. <h3>Materials/Methods</h3> Our institution has set up a newly developed biology-guided radiotherapy (BgRT) system enabling high-quality fan-beam CT imaging for daily treatments. The data of the first prostate patient treated with this system including 26 daily fan-beam CT scans were utilized in this study. A population network was first trained for the auto-segmentation of the prostate and seven pelvic OARs, based on a population dataset containing 56 different patient cases treated on the regular Linac. The pre-trained population network was intentionally fine-tuned to adapt to this specific patient with a transfer learning method. The selected OARs of this patient were contoured by two radiation oncologists on the initial treatment planning and all sets of daily fan-beam CT images as the ground truth for the patient-specific learning. Three different fine-tuning methods, i.e., fine-tuning the last convolution layer, the last convolution block and the whole network were analyzed. A longitudinal study was conducted by exploring the relationship between the auto-segmentation performance and the number of the sequential prior data used for learning. The performance of the patient-specific network was compared with the population network as well as the clinical rigid registration method to highlight its efficiency in radiotherapy practice. <h3>Results</h3> The Dice similarity coefficient (DSC) results of the proposed patient-specific network were 0.882, 0.934, 0.864, 0.827, 0.952, 0.954, 0.921 and 0.904 for the prostate, bladder, rectum, seminal vesicle, left and right femoral head, bowel and lymph nodes, respectively, outperforming the population network (mean DSC of 0.627) and the clinical registration method (mean DSC of 0.716). The whole network fine-tuning method outperformed the other two fine-tuning methods, showing it is more effective in this study. The contouring accuracy of the patient-specific network gradually increased with the increment of longitudinal training cases and approached saturation with more than four training cases. <h3>Conclusion</h3> This study proposes an accurate auto-segmentation method based on the patient-specific deep learning for adaptive prostate radiotherapy. By utilizing the image and contour similarity between the treatment fractions, the patient-specific auto-segmentation could outperform the common population network and the clinical registration method by a large margin, and thus is promising to facilitate more precise radiotherapy based on the new BgRT system.

The clinical evaluation of atlas-based auto-segmentation for automatic contouring during cervical cancer radiotherapy.

PurposeOur purpose was to investigate the influence of atlas library size and CT cross-slice number on the accuracy and efficiency of the atlas-based auto-segmentation (ABAS) method for the automatic contouring of clinical treatment volume (CTV) and organs at risk (OARs) during cervical cancer radiotherapy.MethodsOf 140 cervical cancer patients, contours from 20, 40, 60, 80, 100, and 120 patients were selected incrementally to create six atlas library groups in ABAS. Another 20 tested patients were automatically contoured with the ABAS method and manually contoured by the same professional oncologist. Contours included CTV, bladder, rectum, femoral head-L, femoral head-R, and spinal cord. The CT cross-slice numbers of the 20 tested patients included 61, 65, 72, 75, 81, and 84. The index of dice similarity coefficients (DSCs) and Hausdorff distance (HD) were used to assess the consistency between ABAS automatic contouring and manual contouring. The randomized block analysis of variance and paired t-test were used for statistical analysis.ResultsThe mean DSC values of “CTV, bladder, femoral head, and spinal cord” were all larger than 0.8. The femoral head and spinal cord showed a high degree of agreement between ABAS automatic contouring and manual contouring, with a mean DC >0.80 and HD <1 cm in all atlas library groups. A post-hoc least significant difference comparison indicated that no significant difference had been found between different atlas library sizes with DSC and HD values. For ABAS efficiency, the atlas library size had no effect on the time of ABAS automatic contouring. The time of automatic contouring increased slightly with the increase in CT cross-slice numbers, which were 99.9, 106.8, 114.0, 120.6, 127.9, and 134.8 s with CT cross-slices of 61, 65, 72, 75, 81, and 84, respectively.ConclusionA total of 20 atlas library sizes and a minimum CT cross-slice number including CTV and OARs are enough for ensuring the accuracy and efficiency of ABAS automatic contouring during cervical cancer radiotherapy.

A Novel and Automated Approach to Classify Radiation Induced Lung Tissue Damage on CT Scans.

Simple SummaryRadiation-induced lung damage (RILD) is a common side effect of treating lung cancer with radiotherapy (RT). RILD is visible on CT imaging, and its radiological appearance can vary dramatically from patient to patient as well as across different sub-regions of the lung and treatment volumes. A classification system for RILD able to differentiate radiological damage on a local level would allow us to better understand the underlying patterns of RILD, see how they change over time post irradiation, and link it with clinical outcomes. In this work we propose a five-class morphological lung tissue classification system that can describe parenchymal tissue changes at the voxel level. The classifier was implemented in a fully automated manner using an optimised deep-learning method, then trained and tested using data acquired through a multi-centre clinical trial. The proposed method performed well on an unseen testing dataset. The automated segmentation achieved considerable overlap with manual segmentations (ranging between 26% and 98% for the five classes) and was graded as acceptable by a clinical expert in 88% of cases. This demonstrates it to be suitable for application on a large dataset to help uncover different patterns of changes in the population.Radiation-induced lung damage (RILD) is a common side effect of radiotherapy (RT). The ability to automatically segment, classify, and quantify different types of lung parenchymal change is essential to uncover underlying patterns of RILD and their evolution over time. A RILD dedicated tissue classification system was developed to describe lung parenchymal tissue changes on a voxel-wise level. The classification system was automated for segmentation of five lung tissue classes on computed tomography (CT) scans that described incrementally increasing tissue density, ranging from normal lung (Class 1) to consolidation (Class 5). For ground truth data generation, we employed a two-stage data annotation approach, akin to active learning. Manual segmentation was used to train a stage one auto-segmentation method. These results were manually refined and used to train the stage two auto-segmentation algorithm. The stage two auto-segmentation algorithm was an ensemble of six 2D Unets using different loss functions and numbers of input channels. The development dataset used in this study consisted of 40 cases, each with a pre-radiotherapy, 3-, 6-, 12-, and 24-month follow-up CT scans (n = 200 CT scans). The method was assessed on a hold-out test dataset of 6 cases (n = 30 CT scans). The global Dice score coefficients (DSC) achieved for each tissue class were: Class (1) 99% and 98%, Class (2) 71% and 44%, Class (3) 56% and 26%, Class (4) 79% and 47%, and Class (5) 96% and 92%, for development and test subsets, respectively. The lowest values for the test subsets were caused by imaging artefacts or reflected subgroups that occurred infrequently and with smaller overall parenchymal volumes. We performed qualitative evaluation on the test dataset presenting manual and auto-segmentation to a blinded independent radiologist to rate them as ‘acceptable’, ‘minor disagreement’ or ‘major disagreement’. The auto-segmentation ratings were similar to the manual segmentation, both having approximately 90% of cases rated as acceptable. The proposed framework for auto-segmentation of different lung tissue classes produces acceptable results in the majority of cases and has the potential to facilitate future large studies of RILD.

Automatic 3-D segmentation and volumetric light fluence correction for photoacoustic tomography based on optimal 3-D graph search.

Preclinical imaging with photoacoustic tomography (PAT) has attracted wide attention in recent years since it is capable of providing molecular contrast with deep imaging depth. The automatic extraction and segmentation of the animal in PAT images is crucial for improving image analysis efficiency and enabling advanced image post-processing, such as light fluence (LF) correction for quantitative PAT imaging. Previous automatic segmentation methods are mostly two-dimensional approaches, which failed to conserve the 3-D surface continuity because the image slices were processed separately. This discontinuity problem further hampers LF correction, which, ideally, should be carried out in 3-D due to spatially diffused illumination. Here, to solve these problems, we propose a volumetric auto-segmentation method for small animal PAT imaging based on the 3-D optimal graph search (3-D GS) algorithm. The 3-D GS algorithm takes into account the relation among image slices by constructing a 3-D node-weighted directed graph, and thus ensures surface continuity. In view of the characteristics of PAT images, we improve the original 3-D GS algorithm on graph construction, solution guidance and cost assignment, such that the accuracy and smoothness of the segmented animal surface were guaranteed. We tested the performance of the proposed method by conducting in vivo nude mice imaging experiments with a commercial preclinical cross-sectional PAT system. The results showed that our method successfully retained the continuous global surface structure of the whole 3-D animal body, as well as smooth local subcutaneous tumor boundaries at different development stages. Moreover, based on the 3-D segmentation result, we were able to simulate volumetric LF distribution of the entire animal body and obtained LF corrected PAT images with enhanced structural visibility and uniform image intensity.

P101 - Implementation and clinical validation of an atlas based auto segmentation method of head and neck radiotherapy volumes of interest

P101 - Implementation and clinical validation of an atlas based auto segmentation method of head and neck radiotherapy volumes of interest

A Generalizable Guided Deep Learning Auto-Segmentation Method of Pancreatic GTV on Multi-Protocol Daily MRIs for MR-Guided Adaptive Radiotherapy

Fast and consistent delineation of gross-tumor volume (GTV) on non-contrast enhanced daily MRI is critical but can be challenging in MR-guided Online Adaptive Radiation Therapy (MRgOART). Our previous works have demonstrated the potential of using deep learning (DL) method for pancreatic GTV segmentation on MRI of a single sequence. However, the application of the obtained model is limited by the variations of MRIs from different scan protocols. To overcome the protocol dependence, we propose a generalizable prior knowledge guided DL method for pancreatic GTV auto-segmentation using the guidance of patient-specific contour from the verification MRI (obtained in the treatment room before the first fraction) on subsequent multi-protocol daily MRIs.The daily MRIs acquired from 21 pancreatic cancer patients on a 1.5T MR-Linac during MRgOART were used to demonstrate the method. Two different scan protocols were used: T2 TSE sequence for nine patients and 4D Vane sequence for the remaining twelve patients. Manual contours carefully delineated on each MRI by treating physicians were utilized as ground truth. All the MRIs were pre-processed (e.g., bias corrected, normalized, cropped, and resampled). Each daily MRI was rigidly registered to the verification MRI and a total of 99 image pairs were generated. The DL network backbone used was UNet with 3 encoders interconnected with 3 decoders and a classification block. Model input was the composite of verification MRI, verification GTV mask, and daily MRI. Data augmentation methods such as translation, rotation, flip, guidance-target inversion, and rigid registration error introduction were applied for robust model training. Patient-based leave-one-out-cross-validation was employed and the model performance was evaluated using Dice similarity coefficient (DSC) compared with the ground truth.The average DSC was 0.63 ± 0.18 for all the cases studied. For the cases with GTV > 5cc, the average DSC was 0.76 ± 0.13 and 0.69 ± 0.08 for T2 TSE and 4D Vane MRIs, respectively. The results are reasonably comparable to the inter-observer variation of DSC = 0.71 estimated in a previous study. The time required to generate GTV contour on a daily MRI set was less than 1s on a hardware with an I7 CPU and a GTX 1060 GPU.The proposed prior-knowledge guided deep leaning auto-segmentation method for pancreatic GTV on daily multi-protocol non-contrast MRIs performed reasonably well even with the relatively small dataset. This proof-of-principle study suggests that, with patient-specific prior knowledge guidance, it is feasible to implement a single DL auto-segmentation model for multiple MRI sequences in MRgOART. The method can be generalized for other structures or tumors. (First and second authors contributed equally.).

A novel semi auto-segmentation method for accurate dose and NTCP evaluation in adaptive head and neck radiotherapy

Background and purposeAccurate segmentation of organs-at-risk (OARs) is crucial but tedious and time-consuming in adaptive radiotherapy (ART). The purpose of this work was to automate head and neck OAR-segmentation on repeat CT (rCT) by an optimal combination of human and auto-segmentation for accurate prediction of Normal Tissue Complication Probability (NTCP). Materials and methodsHuman segmentation (HS) of 3 observers, deformable image registration (DIR) based contour propagation and deep learning contouring (DLC) were carried out to segment 15 OARs on 15 rCTs. The original treatment plan was re-calculated on rCT to obtain mean dose (Dmean) and consequent NTCP-predictions. The average Dmean and NTCP-predictions of the three observers were referred to as the gold standard to calculate the absolute difference of Dmean and NTCP-predictions (|ΔDmean| and |ΔNTCP|). ResultsThe average |ΔDmean| of parotid glands in HS was 1.40 Gy, lower than that obtained with DIR and DLC (3.64 Gy, p < 0.001 and 3.72 Gy, p < 0.001, respectively). DLC showed the highest |ΔDmean| in middle Pharyngeal Constrictor Muscle (PCM) (5.13 Gy, p = 0.01). DIR showed second highest |ΔDmean| in the cricopharyngeal inlet (2.85 Gy, p = 0.01). The semi auto-segmentation (SAS) adopted HS, DIR and DLC for segmentation of parotid glands, PCM and all other OARs, respectively. The 90th percentile |ΔNTCP|was 2.19%, 2.24%, 1.10% and 1.50% for DIR, DLC, HS and SAS respectively. ConclusionsHuman segmentation of the parotid glands remains necessary for accurate interpretation of mean dose and NTCP during ART. Proposed semi auto-segmentation allows NTCP-predictions within 1.5% accuracy for 90% of the cases.

PO-1686 A novel semi auto-segmentation method for head and neck adaptive radiotherapy

PO-1686 A novel semi auto-segmentation method for head and neck adaptive radiotherapy