A prior-information-based automatic segmentation method for the clinical target volume in adaptive radiotherapy of cervical cancer

OBJECTIVE

Adaptive planning to accommodate anatomic changes during treatment often requires repeated segmentation. In this study, prior patient-specific data was integrateda into a registration-guided multi-channel multi-path (Rg-MCMP) segmentation framework to improve the accuracy of repeated clinical target volume (CTV) segmentation.

METHODS

This study was based on CT image datasets for a total of 90 cervical cancer patients who received two courses of radiotherapy. A total of 15 patients were selected randomly as the test set. In the Rg-MCMP segmentation framework, the first-course CT images (CT1) were registered to second-course CT images (CT2) to yield aligned CT images (aCT1), and the CTV in the first course (CTV1) was propagated to yield aligned CTV contours (aCTV1). Then, aCT1, aCTV1, and CT2 were combined as the inputs for 3D U-Net consisting of a channel-based multi-path feature extraction network. The performance of the Rg-MCMP segmentation framework was evaluated and compared with the single-channel single-path model (SCSP), the standalone registration methods, and the registration-guided multi-channel single-path (Rg-MCSP) model. The Dice similarity coefficient (DSC), 95% Hausdorff distance (HD95), and average surface distance (ASD) were used as the metrics.

RESULTS

The average DSC of CTV for the deformable image DIR-MCMP model was found to be 0.892, greater than that of the standalone DIR (0.856), SCSP (0.837), and DIR-MCSP (0.877), which were improvements of 4.2%, 6.6%, and 1.7%, respectively. Similarly, the rigid body DIR-MCMP model yielded an average DSC of 0.875, which exceeded standalone RB (0.787), SCSP (0.837), and registration-guided multi-channel single-path (0.848), which were improvements of 11.2%, 4.5%, and 3.2%, respectively. These improvements in DSC were statistically significant (p < 0.05).

CONCLUSION

The proposed Rg-MCMP framework achieved excellent accuracy in CTV segmentation as part of the adaptive radiotherapy workflow.

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

BACKGROUND AND PURPOSE

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.

METHODS

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.

RESULTS

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.

CONCLUSION

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.

Low-field magnetic resonance image enhancement via stochastic image quality transfer

Low-field (<1T) magnetic resonance imaging (MRI) scanners remain in widespread use in low- and middle-income countries (LMICs) and are commonly used for some applications in higher income countries e.g. for small child patients with obesity, claustrophobia, implants, or tattoos. However, low-field MR images commonly have lower resolution and poorer contrast than images from high field (1.5T, 3T, and above). Here, we present Image Quality Transfer (IQT) to enhance low-field structural MRI by estimating from a low-field image the image we would have obtained from the same subject at high field. Our approach uses (i) a stochastic low-field image simulator as the forward model to capture uncertainty and variation in the contrast of low-field images corresponding to a particular high-field image, and (ii) an anisotropic U-Net variant specifically designed for the IQT inverse problem. We evaluate the proposed algorithm both in simulation and using multi-contrast (T1-weighted, T2-weighted, and fluid attenuated inversion recovery (FLAIR)) clinical low-field MRI data from an LMIC hospital. We show the efficacy of IQT in improving contrast and resolution of low-field MR images. We demonstrate that IQT-enhanced images have potential for enhancing visualisation of anatomical structures and pathological lesions of clinical relevance from the perspective of radiologists. IQT is proved to have capability of boosting the diagnostic value of low-field MRI, especially in low-resource settings.

Improved accuracy of auto-segmentation of organs at risk in radiotherapy planning for nasopharyngeal carcinoma based on fully convolutional neural network deep learning

OBJECTIVE

We examined a modified encoder-decoder architecture-based fully convolutional neural network, OrganNet, for simultaneous auto-segmentation of 24 organs at risk (OARs) in the head and neck, followed by validation tests and evaluation of clinical application.

MATERIALS AND METHODS

Computed tomography (CT) images from 310 radiotherapy plans were used as the experimental data set, of which 260 and 50 were used as the training and test sets, respectively. An improved U-Net architecture was established by introducing a batch normalization layer, residual squeeze-and-excitation layer, and unique organ-specific loss function for deep learning training. The performance of the trained network model was evaluated by comparing the manual-delineation and the STAPLE contour of 10 physicians from different centers.

RESULTS

Our model achieved good segmentation in all 24 OARs in nasopharyngeal cancer radiotherapy plan CT images, with an average Dice similarity coefficient of 83.75%. Specifically, the mean Dice coefficients in large-volume organs (brainstem, spinal cord, left/right parotid glands, left/right temporal lobes, and left/right mandibles) were 84.97% - 95.00%, and in small-volume organs (pituitary, lens, optic nerve, and optic chiasma) were 55.46% - 91.56%. respectively. Using the STAPLE contours as standard contour, the OrganNet achieved comparable or better DICE in organ segmentation then that of the manual-delineation as well.

CONCLUSION

The established OrganNet enables simultaneous automatic segmentation of multiple targets on CT images of the head and neck radiotherapy plans, effectively improves the accuracy of U-Net based segmentation for OARs, especially for small-volume organs.

Automatic Delineation of Gross Tumor Volume Based on Magnetic Resonance Imaging by Performing a Novel Semisupervised Learning Framework in Nasopharyngeal Carcinoma

PURPOSE

We aimed to validate the accuracy and clinical value of a novel semisupervised learning framework for gross tumor volume (GTV) delineation in nasopharyngeal carcinoma.

METHODS AND MATERIALS

Two hundred fifty-eight patients with magnetic resonance imaging data sets were divided into training (n = 180), validation (n = 20), and testing (n = 58) cohorts. Ground truth contours of nasopharynx GTV (GTVnx) and node GTV (GTVnd) were manually delineated by 2 experienced radiation oncologists. Twenty percent (n = 36) labeled and 80% (n = 144) unlabeled images were used to train the model, producing model-generated contours for patients from the testing cohort. Nine experienced experts were invited to revise model-generated GTV in 20 randomly selected patients from the testing cohort. Six junior oncologists were asked to delineate GTV in 12 randomly selected patients from the testing cohort without and with the assistance of the model, and revision degrees were compared under these 2 modes. The Dice similarity coefficient (DSC) was used to quantify the accuracy of the model.

RESULTS

The model-generated contours showed a high accuracy compared with ground truth contours, with an average DSC score of 0.83 and 0.80 for GTVnx and GTVnd, respectively. There was no significant difference in DSC score between T1-2 and T3-4 patients (0.81 vs 0.83; P = .223), or between N1-2 and N3 patients (0.80 vs 0.79; P = .807). The mean revision degree was lower than 10% in 19 (95%) patients for GTVnx and in 16 (80%) patients for GTVnd. With assistance of the model, the mean revision degree for GTVnx and GTVnd by junior oncologists was reduced from 25.63% to 7.75% and from 21.38% to 14.44%, respectively. Meanwhile, the delineating efficiency was improved by over 60%.

CONCLUSIONS

The proposed semisupervised learning-based model showed a high accuracy for delineating GTV of nasopharyngeal carcinoma. It was clinically applicable and could assist junior oncologists to improve GTV contouring accuracy and save contouring time.

Clinical evaluation of deep learning-based clinical target volume three-channel auto-segmentation algorithm for adaptive radiotherapy in cervical cancer

Objectives

Accurate contouring of the clinical target volume (CTV) is a key element of radiotherapy in cervical cancer. We validated a novel deep learning (DL)-based auto-segmentation algorithm for CTVs in cervical cancer called the three-channel adaptive auto-segmentation network (TCAS).

Methods

A total of 107 cases were collected and contoured by senior radiation oncologists (ROs). Each case consisted of the following: (1) contrast-enhanced CT scan for positioning, (2) the related CTV, (3) multiple plain CT scans during treatment and (4) the related CTV. After registration between (1) and (3) for the same patient, the aligned image and CTV were generated. Method 1 is rigid registration, method 2 is deformable registration, and the aligned CTV is seen as the result. Method 3 is rigid registration and TCAS, method 4 is deformable registration and TCAS, and the result is generated by a DL-based method.

Results

From the 107 cases, 15 pairs were selected as the test set. The dice similarity coefficient (DSC) of method 1 was 0.8155 ± 0.0368; the DSC of method 2 was 0.8277 ± 0.0315; the DSCs of method 3 and 4 were 0.8914 ± 0.0294 and 0.8921 ± 0.0231, respectively. The mean surface distance and Hausdorff distance of methods 3 and 4 were markedly better than those of method 1 and 2.

Conclusions

The TCAS achieved comparable accuracy to the manual delineation performed by senior ROs and was significantly better than direct registration.

Improved Otsu and Kapur approach for white blood cells segmentation based on LebTLBO optimization for the detection of Leukemia

The diagnosis of leukemia involves the detection of the abnormal characteristics of blood cells by a trained pathologist. Currently, this is done manually by observing the morphological characteristics of white blood cells in the microscopic images. Though there are some equipment- based and chemical-based tests available, the use and adaptation of the automated computer vision-based system is still an issue. There are certain software frameworks available in the literature; however, they are still not being adopted commercially. So there is a need for an automated and software- based framework for the detection of leukemia. In software-based detection, segmentation is the first critical stage that outputs the region of interest for further accurate diagnosis. Therefore, this paper explores an efficient and hybrid segmentation that proposes a more efficient and effective system for leukemia diagnosis. A very popular publicly available database, the acute lymphoblastic leukemia image database (ALL-IDB), is used in this research. First, the images are pre-processed and segmentation is done using Multilevel thresholding with Otsu and Kapur methods. To further optimize the segmentation performance, the Learning enthusiasm-based teaching-learning-based optimization (LebTLBO) algorithm is employed. Different metrics are used for measuring the system performance. A comparative analysis of the proposed methodology is done with existing benchmarks methods. The proposed approach has proven to be better than earlier techniques with measuring parameters of PSNR and Similarity index. The result shows a significant improvement in the performance measures with optimizing threshold algorithms and the LebTLBO technique.

Deep learning-based auto-segmentation of clinical target volumes for radiotherapy treatment of cervical cancer

Objectives

Because radiotherapy is indispensible for treating cervical cancer, it is critical to accurately and efficiently delineate the radiation targets. We evaluated a deep learning (DL)‐based auto‐segmentation algorithm for automatic contouring of clinical target volumes (CTVs) in cervical cancers.

Methods

Computed tomography (CT) datasets from 535 cervical cancers treated with definitive or postoperative radiotherapy were collected. A DL tool based on VB‐Net was developed to delineate CTVs of the pelvic lymph drainage area (dCTV1) and parametrial area (dCTV2) in the definitive radiotherapy group. The training/validation/test number is 157/20/23. CTV of the pelvic lymph drainage area (pCTV1) was delineated in the postoperative radiotherapy group. The training/validation/test number is 272/30/33. Dice similarity coefficient (DSC), mean surface distance (MSD), and Hausdorff distance (HD) were used to evaluate the contouring accuracy. Contouring times were recorded for efficiency comparison.

Results

The mean DSC, MSD, and HD values for our DL‐based tool were 0.88/1.32 mm/21.60 mm for dCTV1, 0.70/2.42 mm/22.44 mm for dCTV2, and 0.86/1.15 mm/20.78 mm for pCTV1. Only minor modifications were needed for 63.5% of auto‐segmentations to meet the clinical requirements. The contouring accuracy of the DL‐based tool was comparable to that of senior radiation oncologists and was superior to that of junior/intermediate radiation oncologists. Additionally, DL assistance improved the performance of junior radiation oncologists for dCTV2 and pCTV1 contouring (mean DSC increases: 0.20 for dCTV2, 0.03 for pCTV1; mean contouring time decrease: 9.8 min for dCTV2, 28.9 min for pCTV1).

Conclusions

DL‐based auto‐segmentation improves CTV contouring accuracy, reduces contouring time, and improves clinical efficiency for treating cervical cancer.

High-order Feature Learning for Multi-atlas based Label Fusion: Application to Brain Segmentation with MRI

Multi-atlas based segmentation methods have shown their effectiveness in brain regions-of-interesting (ROIs) segmentation, by propagating labels from multiple atlases to a target image based on the similarity between patches in the target image and multiple atlas images. Most of the existing multiatlas based methods use image intensity features to calculate the similarity between a pair of image patches for label fusion. In particular, using only low-level image intensity features cannot adequately characterize the complex appearance patterns (e.g., the high-order relationship between voxels within a patch) of brain magnetic resonance (MR) images. To address this issue, this paper develops a high-order feature learning framework for multi-atlas based label fusion, where high-order features of image patches are extracted and fused for segmenting ROIs of structural brain MR images. Specifically, an unsupervised feature learning method (i.e., means-covariances restricted Boltzmann machine, mcRBM) is employed to learn high-order features (i.e., mean and covariance features) of patches in brain MR images. Then, a group-fused sparsity dictionary learning method is proposed to jointly calculate the voting weights for label fusion, based on the learned high-order and the original image intensity features. The proposed method is compared with several state-of-the-art label fusion methods on ADNI, NIREP and LONI-LPBA40 datasets. The Dice ratio achieved by our method is 88:30%, 88:83%, 79:54% and 81:02% on left and right hippocampus on the ADNI, NIREP and LONI-LPBA40 datasets, respectively, while the best Dice ratio yielded by the other methods are 86:51%, 87:39%, 78:48% and 79:65% on three datasets, respectively.

Quantifying deep grey matter atrophy using automated segmentation approaches: A systematic review of structural MRI studies

The deep grey matter (DGM) nuclei of the brain play a crucial role in learning, behaviour, cognition, movement and memory. Although automated segmentation strategies can provide insight into the impact of multiple neurological conditions affecting these structures, such as Multiple Sclerosis (MS), Huntington's disease (HD), Alzheimer's disease (AD), Parkinson's disease (PD) and Cerebral Palsy (CP), there are a number of technical challenges limiting an accurate automated segmentation of the DGM. Namely, the insufficient contrast of T1 sequences to completely identify the boundaries of these structures, as well as the presence of iso-intense white matter lesions or extensive tissue loss caused by brain injury. Therefore in this systematic review, 269 eligible studies were analysed and compared to determine the optimal approaches for addressing these technical challenges. The automated approaches used among the reviewed studies fall into three broad categories, atlas-based approaches focusing on the accurate alignment of atlas priors, algorithmic approaches which utilise intensity information to a greater extent, and learning-based approaches that require an annotated training set. Studies that utilise freely available software packages such as FIRST, FreeSurfer and LesionTOADS were also eligible, and their performance compared. Overall, deep learning approaches achieved the best overall performance, however these strategies are currently hampered by the lack of large-scale annotated data. Improving model generalisability to new datasets could be achieved in future studies with data augmentation and transfer learning. Multi-atlas approaches provided the second-best performance overall, and may be utilised to construct a "silver standard" annotated training set for deep learning. To address the technical challenges, providing robustness to injury can be improved by using multiple channels, highly elastic diffeomorphic transformations such as LDDMM, and by following atlas-based approaches with an intensity driven refinement of the segmentation, which has been done with the Expectation Maximisation (EM) and level sets methods. Accounting for potential lesions should be achieved with a separate lesion segmentation approach, as in LesionTOADS. Finally, to address the issue of limited contrast, R2*, T2* and QSM sequences could be used to better highlight the DGM due to its higher iron content. Future studies could look to additionally acquire these sequences by retaining the phase information from standard structural scans, or alternatively acquiring these sequences for only a training set, allowing models to learn the "improved" segmentation from T1-sequences alone.

Deep Learning for Automated Contouring of Primary Tumor Volumes by MRI for Nasopharyngeal Carcinoma

Background Nasopharyngeal carcinoma (NPC) may be cured with radiation therapy. Tumor proximity to critical structures demands accuracy in tumor delineation to avoid toxicities from radiation therapy; however, tumor target contouring for head and neck radiation therapy is labor intensive and highly variable among radiation oncologists. Purpose To construct and validate an artificial intelligence (AI) contouring tool to automate primary gross tumor volume (GTV) contouring in patients with NPC. Materials and Methods In this retrospective study, MRI data sets covering the nasopharynx from 1021 patients (median age, 47 years; 751 male, 270 female) with NPC between September 2016 and September 2017 were collected and divided into training, validation, and testing cohorts of 715, 103, and 203 patients, respectively. GTV contours were delineated for 1021 patients and were defined by consensus of two experts. A three-dimensional convolutional neural network was applied to 818 training and validation MRI data sets to construct the AI tool, which was tested in 203 independent MRI data sets. Next, the AI tool was compared against eight qualified radiation oncologists in a multicenter evaluation by using a random sample of 20 test MRI examinations. The Wilcoxon matched-pairs signed rank test was used to compare the difference of Dice similarity coefficient (DSC) of pre- versus post-AI assistance. Results The AI-generated contours demonstrated a high level of accuracy when compared with ground truth contours at testing in 203 patients (DSC, 0.79; 2.0-mm difference in average surface distance). In multicenter evaluation, AI assistance improved contouring accuracy (five of eight oncologists had a higher median DSC after AI assistance; average median DSC, 0.74 vs 0.78; P < .001), reduced intra- and interobserver variation (by 36.4% and 54.5%, respectively), and reduced contouring time (by 39.4%). Conclusion The AI contouring tool improved primary gross tumor contouring accuracy of nasopharyngeal carcinoma, which could have a positive impact on tumor control and patient survival. © RSNA, 2019 Online supplemental material is available for this article. See also the editorial by Chang in this issue.

7T MRI subthalamic nucleus atlas for use with 3T MRI

Deep brain stimulation (DBS) of the subthalamic nucleus (STN) reduces motor symptoms in most patients with Parkinson disease (PD), yet may produce untoward effects. Investigation of DBS effects requires accurate localization of the STN, which can be difficult to identify on magnetic resonance images collected with clinically available 3T scanners. The goal of this study is to develop a high-quality STN atlas that can be applied to standard 3T images. We created a high-definition STN atlas derived from seven older participants imaged at 7T. This atlas was nonlinearly registered to a standard template representing 56 patients with PD imaged at 3T. This process required development of methodology for nonlinear multimodal image registration. We demonstrate mm-scale STN localization accuracy by comparison of our 3T atlas with a publicly available 7T atlas. We also demonstrate less agreement with an earlier histological atlas. STN localization error in the 56 patients imaged at 3T was less than 1 mm on average. Our methodology enables accurate STN localization in individuals imaged at 3T. The STN atlas and underlying 3T average template in MNI space are freely available to the research community. The image registration methodology developed in the course of this work may be generally applicable to other datasets.

Discrete pre-processing step effects in registration-based pipelines, a preliminary volumetric study on T1-weighted images

Pre-processing MRI scans prior to performing volumetric analyses is common practice in MRI studies. As pre-processing steps adjust the voxel intensities, the space in which the scan exists, and the amount of data in the scan, it is possible that the steps have an effect on the volumetric output. To date, studies have compared between and not within pipelines, and so the impact of each step is unknown. This study aims to quantify the effects of pre-processing steps on volumetric measures in T1-weighted scans within a single pipeline. It was our hypothesis that pre-processing steps would significantly impact ROI volume estimations. One hundred fifteen participants from the OASIS dataset were used, where each participant contributed three scans. All scans were then pre-processed using a step-wise pipeline. Bilateral hippocampus, putamen, and middle temporal gyrus volume estimations were assessed following each successive step, and all data were processed by the same pipeline 5 times. Repeated-measures analyses tested for a main effects of pipeline step, scan-rescan (for MRI scanner consistency) and repeated pipeline runs (for algorithmic consistency). A main effect of pipeline step was detected, and interestingly an interaction between pipeline step and ROI exists. No effect for either scan-rescan or repeated pipeline run was detected. We then supply a correction for noise in the data resulting from pre-processing.

Development and Implementation of a Corriedale Ovine Brain Atlas for Use in Atlas-Based Segmentation

Segmentation is the process of partitioning an image into subdivisions and can be applied to medical images to isolate anatomical or pathological areas for further analysis. This process can be done manually or automated by the use of image processing computer packages. Atlas-based segmentation automates this process by the use of a pre-labelled template and a registration algorithm. We developed an ovine brain atlas that can be used as a model for neurological conditions such as Parkinson's disease and focal epilepsy. 17 female Corriedale ovine brains were imaged in-vivo in a 1.5T (low-resolution) MRI scanner. 13 of the low-resolution images were combined using a template construction algorithm to form a low-resolution template. The template was labelled to form an atlas and tested by comparing manual with atlas-based segmentations against the remaining four low-resolution images. The comparisons were in the form of similarity metrics used in previous segmentation research. Dice Similarity Coefficients were utilised to determine the degree of overlap between eight independent, manual and atlas-based segmentations, with values ranging from 0 (no overlap) to 1 (complete overlap). For 7 of these 8 segmented areas, we achieved a Dice Similarity Coefficient of 0.5-0.8. The amygdala was difficult to segment due to its variable location and similar intensity to surrounding tissues resulting in Dice Coefficients of 0.0-0.2. We developed a low resolution ovine brain atlas with eight clinically relevant areas labelled. This brain atlas performed comparably to prior human atlases described in the literature and to intra-observer error providing an atlas that can be used to guide further research using ovine brains as a model and is hosted online for public access.

Peripheral nerve segmentation using Nonparametric Bayesian Hierarchical Clustering

Several cases related to chronic pain, due to accidents, illness or surgical interventions, depend on anesthesiology procedures. These procedures are assisted with ultrasound images. Although, the ultrasound images are a useful instrument in order to guide the specialist in anesthesiology, the lack of intelligibility due to speckle noise, makes the clinical intervention a difficult task. In a similar manner, some artifacts are introduced in the image capturing process, challenging the expertise of anesthesiologists for not confusing the true nerve structures. Accordingly, an assistance methodology using image processing can improve the accuracy in the anesthesia practice. This paper proposes a peripheral nerve segmentation method in medical ultrasound images, based on Nonparametric Bayesian Hierarchical Clustering. The experimental results show segmentation performances with a Mean Squared Error performance of 1.026 ± 0.379 pixels for ulnar nerve, 0.704 ± 0.233 pixels for median nerve and 1.698 ± 0.564 pixels for peroneal nerve. Likewise, the model allows to emphasize other soft structures like muscles and aqueous tissues, that might be useful for an anesthesiologist.

Alpha shape theory for 3D visualization and volumetric measurement of brain tumor progression using magnetic resonance images

Resection of brain tumors is a tricky task in surgery due to its direct influence on the patients' survival rate. Determining the tumor resection extent for its complete information via-à-vis volume and dimensions in pre- and post-operative Magnetic Resonance Images (MRI) requires accurate estimation and comparison. The active contour segmentation technique is used to segment brain tumors on pre-operative MR images using self-developed software. Tumor volume is acquired from its contours via alpha shape theory. The graphical user interface is developed for rendering, visualizing and estimating the volume of a brain tumor. Internet Brain Segmentation Repository dataset (IBSR) is employed to analyze and determine the repeatability and reproducibility of tumor volume. Accuracy of the method is validated by comparing the estimated volume using the proposed method with that of gold-standard. Segmentation by active contour technique is found to be capable of detecting the brain tumor boundaries. Furthermore, the volume description and visualization enable an interactive examination of tumor tissue and its surrounding. Admirable features of our results demonstrate that alpha shape theory in comparison to other existing standard methods is superior for precise volumetric measurement of tumor.

A direct morphometric comparison of five labeling protocols for multi-atlas driven automatic segmentation of the hippocampus in Alzheimer's disease

Hippocampal volumetry derived from structural MRI is increasingly used to delineate regions of interest for functional measurements, assess efficacy in therapeutic trials of Alzheimer's disease (AD) and has been endorsed by the new AD diagnostic guidelines as a radiological marker of disease progression. Unfortunately, morphological heterogeneity in AD can prevent accurate demarcation of the hippocampus. Recent developments in automated volumetry commonly use multi-template fusion driven by expert manual labels, enabling highly accurate and reproducible segmentation in disease and healthy subjects. However, there are several protocols to define the hippocampus anatomically in vivo, and the method used to generate atlases may impact automatic accuracy and sensitivity - particularly in pathologically heterogeneous samples. Here we report a fully automated segmentation technique that provides a robust platform to directly evaluate both technical and biomarker performance in AD among anatomically unique labeling protocols. For the first time we test head-to-head the performance of five common hippocampal labeling protocols for multi-atlas based segmentation, using both the Sunnybrook Longitudinal Dementia Study and the entire Alzheimer's Disease Neuroimaging Initiative 1 (ADNI-1) baseline and 24-month dataset. We based these atlas libraries on the protocols of (Haller et al., 1997; Killiany et al., 1993; Malykhin et al., 2007; Pantel et al., 2000; Pruessner et al., 2000), and a single operator performed all manual tracings to generate de facto "ground truth" labels. All methods distinguished between normal elders, mild cognitive impairment (MCI), and AD in the expected directions, and showed comparable correlations with measures of episodic memory performance. Only more inclusive protocols distinguished between stable MCI and MCI-to-AD converters, and had slightly better associations with episodic memory. Moreover, we demonstrate that protocols including more posterior anatomy and dorsal white matter compartments furnish the best voxel-overlap accuracies (Dice Similarity Coefficient=0.87-0.89), compared to expert manual tracings, and achieve the smallest sample sizes required to power clinical trials in MCI and AD. The greatest distribution of errors was localized to the caudal hippocampus and the alveus-fimbria compartment when these regions were excluded. The definition of the medial body did not significantly alter accuracy among more comprehensive protocols. Voxel-overlap accuracies between automatic and manual labels were lower for the more pathologically heterogeneous Sunnybrook study in comparison to the ADNI-1 sample. Finally, accuracy among protocols appears to significantly differ the most in AD subjects compared to MCI and normal elders. Together, these results suggest that selection of a candidate protocol for fully automatic multi-template based segmentation in AD can influence both segmentation accuracy when compared to expert manual labels and performance as a biomarker in MCI and AD.