Minimizing the effect of white matter lesions on deep learning based tissue segmentation for brain volumetry.

Automated methods for segmentation-based brain volumetry may be confounded by the presence of white matter (WM) lesions, which introduce abnormal intensities that can alter the classification of not only neighboring but also distant brain tissue. These lesions are common in pathologies where brain volumetry is also an important prognostic marker, such as in multiple sclerosis (MS), and thus reducing their effects is critical for improving volumetric accuracy and reliability. In this work, we analyze the effect of WM lesions on deep learning based brain tissue segmentation methods for brain volumetry and introduce techniques to reduce the error these lesions produce on the measured volumes. We propose a 3D patch-based deep learning framework for brain tissue segmentation which is trained on the outputs of a reference classical method. To deal more robustly with pathological cases having WM lesions, we use a combination of small patches and a percentile-based input normalization. To minimize the effect of WM lesions, we also propose a multi-task double U-Net architecture performing end-to-end inpainting and segmentation, along with a training data generation procedure. In the evaluation, we first analyze the error introduced by artificial WM lesions on our framework as well as in the reference segmentation method without the use of lesion inpainting techniques. To the best of our knowledge, this is the first analysis of WM lesion effect on a deep learning based tissue segmentation approach for brain volumetry. The proposed framework shows a significantly smaller and more localized error introduced by WM lesions than the reference segmentation method, that displays much larger global differences. We also evaluated the proposed lesion effect minimization technique by comparing the measured volumes before and after introducing artificial WM lesions to healthy images. The proposed approach performing end-to-end inpainting and segmentation effectively reduces the error introduced by small and large WM lesions in the resulting volumetry, obtaining absolute volume differences of 0.01 ± 0.03% for GM and 0.02 ± 0.04% for WM. Increasing the accuracy and reliability of automated brain volumetry methods will reduce the sample size needed to establish meaningful correlations in clinical studies and allow its use in individualized assessments as a diagnostic and prognostic marker for neurodegenerative pathologies.

Transductive Transfer Learning for Domain Adaptation in Brain Magnetic Resonance Image Segmentation.

Segmentation of brain images from Magnetic Resonance Images (MRI) is an indispensable step in clinical practice. Morphological changes of sub-cortical brain structures and quantification of brain lesions are considered biomarkers of neurological and neurodegenerative disorders and used for diagnosis, treatment planning, and monitoring disease progression. In recent years, deep learning methods showed an outstanding performance in medical image segmentation. However, these methods suffer from generalisability problem due to inter-centre and inter-scanner variabilities of the MRI images. The main objective of the study is to develop an automated deep learning segmentation approach that is accurate and robust to the variabilities in scanner and acquisition protocols. In this paper, we propose a transductive transfer learning approach for domain adaptation to reduce the domain-shift effect in brain MRI segmentation. The transductive scenario assumes that there are sets of images from two different domains: (1) source-images with manually annotated labels; and (2) target-images without expert annotations. Then, the network is jointly optimised integrating both source and target images into the transductive training process to segment the regions of interest and to minimise the domain-shift effect. We proposed to use a histogram loss in the feature level to carry out the latter optimisation problem. In order to demonstrate the benefit of the proposed approach, the method has been tested in two different brain MRI image segmentation problems using multi-centre and multi-scanner databases for: (1) sub-cortical brain structure segmentation; and (2) white matter hyperintensities segmentation. The experiments showed that the segmentation performance of a pre-trained model could be significantly improved by up to 10%. For the first segmentation problem it was possible to achieve a maximum improvement from 0.680 to 0.799 in average Dice Similarity Coefficient (DSC) metric and for the second problem the average DSC improved from 0.504 to 0.602. Moreover, the improvements after domain adaptation were on par or showed better performance compared to the commonly used traditional unsupervised segmentation methods (FIRST and LST), also achieving faster execution time. Taking this into account, this work presents one more step toward the practical implementation of deep learning algorithms into the clinical routine.

A fully convolutional neural network for new T2-w lesion detection in multiple sclerosis.

INTRODUCTION: Longitudinal magnetic resonance imaging (MRI) has an important role in multiple sclerosis (MS) diagnosis and follow-up. Specifically, the presence of new T2-w lesions on brain MR scans is considered a predictive biomarker for the disease. In this study, we propose a fully convolutional neural network (FCNN) to detect new T2-w lesions in longitudinal brain MR images. METHODS: One year apart, multichannel brain MR scans (T1-w, T2-w, PD-w, and FLAIR) were obtained for 60 patients, 36 of them with new T2-w lesions. Modalities from both temporal points were preprocessed and linearly coregistered. Afterwards, an FCNN, whose inputs were from the baseline and follow-up images, was trained to detect new MS lesions. The first part of the network consisted of U-Net blocks that learned the deformation fields (DFs) and nonlinearly registered the baseline image to the follow-up image for each input modality. The learned DFs together with the baseline and follow-up images were then fed to the second part, another U-Net that performed the final detection and segmentation of new T2-w lesions. The model was trained end-to-end, simultaneously learning both the DFs and the new T2-w lesions, using a combined loss function. We evaluated the performance of the model following a leave-one-out cross-validation scheme. RESULTS: In terms of the detection of new lesions, we obtained a mean Dice similarity coefficient of 0.83 with a true positive rate of 83.09% and a false positive detection rate of 9.36%. In terms of segmentation, we obtained a mean Dice similarity coefficient of 0.55. The performance of our model was significantly better compared to the state-of-the-art methods (p < 0.05). CONCLUSIONS: Our proposal shows the benefits of combining a learning-based registration network with a segmentation network. Compared to other methods, the proposed model decreases the number of false positives. During testing, the proposed model operates faster than the other two state-of-the-art methods based on the DF obtained by Demons.

One-shot domain adaptation in multiple sclerosis lesion segmentation using convolutional neural networks.

In recent years, several convolutional neural network (CNN) methods have been proposed for the automated white matter lesion segmentation of multiple sclerosis (MS) patient images, due to their superior performance compared with those of other state-of-the-art methods. However, the accuracies of CNN methods tend to decrease significantly when evaluated on different image domains compared with those used for training, which demonstrates the lack of adaptability of CNNs to unseen imaging data. In this study, we analyzed the effect of intensity domain adaptation on our recently proposed CNN-based MS lesion segmentation method. Given a source model trained on two public MS datasets, we investigated the transferability of the CNN model when applied to other MRI scanners and protocols, evaluating the minimum number of annotated images needed from the new domain and the minimum number of layers needed to re-train to obtain comparable accuracy. Our analysis comprised MS patient data from both a clinical center and the public ISBI2015 challenge database, which permitted us to compare the domain adaptation capability of our model to that of other state-of-the-art methods. In both datasets, our results showed the effectiveness of the proposed model in adapting previously acquired knowledge to new image domains, even when a reduced number of training samples was available in the target dataset. For the ISBI2015 challenge, our one-shot domain adaptation model trained using only a single case showed a performance similar to that of other CNN methods that were fully trained using the entire available training set, yielding a comparable human expert rater performance. We believe that our experiments will encourage the MS community to incorporate its use in different clinical settings with reduced amounts of annotated data. This approach could be meaningful not only in terms of the accuracy in delineating MS lesions but also in the related reductions in time and economic costs derived from manual lesion labeling.

A supervised framework with intensity subtraction and deformation field features for the detection of new T2-w lesions in multiple sclerosis.

INTRODUCTION: Longitudinal magnetic resonance imaging (MRI) analysis has an important role in multiple sclerosis diagnosis and follow-up. The presence of new T2-w lesions on brain MRI scans is considered a prognostic and predictive biomarker for the disease. In this study, we propose a supervised approach for detecting new T2-w lesions using features from image intensities, subtraction values, and deformation fields (DF). METHODS: One year apart multi-channel brain MRI scans were obtained for 60 patients, 36 of them with new T2-w lesions. Images from both temporal points were preprocessed and co-registered. Afterwards, they were registered using multi-resolution affine registration, allowing their subtraction. In particular, the DFs between both images were computed with the Demons non-rigid registration algorithm. Afterwards, a logistic regression model was trained with features from image intensities, subtraction values, and DF operators. We evaluated the performance of the model following a leave-one-out cross-validation scheme. RESULTS: In terms of detection, we obtained a mean Dice similarity coefficient of 0.77 with a true-positive rate of 74.30% and a false-positive detection rate of 11.86%. In terms of segmentation, we obtained a mean Dice similarity coefficient of 0.56. The performance of our model was significantly higher than state-of-the-art methods. CONCLUSIONS: The performance of the proposed method shows the benefits of using DF operators as features to train a supervised learning model. Compared to other methods, the proposed model decreases the number of false-positives while increasing the number of true-positives, which is relevant for clinical settings.

Dynamic 3D shape of the plantar surface of the foot using coded structured light: a technical report.

BACKGROUND: The foot provides a crucial contribution to the balance and stability of the musculoskeletal system, and accurate foot measurements are important in applications such as designing custom insoles/footwear. With better understanding of the dynamic behavior of the foot, dynamic foot reconstruction techniques are surfacing as useful ways to properly measure the shape of the foot. This paper presents a novel design and implementation of a structured-light prototype system providing dense three dimensional (3D) measurements of the foot in motion. The input to the system is a video sequence of a foot during a single step; the output is a 3D reconstruction of the plantar surface of the foot for each frame of the input. METHODS: Engineering and clinical tests were carried out to test the accuracy and repeatability of the system. Accuracy experiments involved imaging a planar surface from different orientations and elevations and measuring the fitting errors of the data to a plane. Repeatability experiments were done using reconstructions from 27 different subjects, where for each one both right and left feet were reconstructed in static and dynamic conditions over two different days. RESULTS: The static accuracy of the system was found to be 0.3 mm with planar test objects. In tests with real feet, the system proved repeatable, with reconstruction differences between trials one week apart averaging 2.4 mm (static case) and 2.8 mm (dynamic case). CONCLUSION: The results obtained in the experiments show positive accuracy and repeatability results when compared to current literature. The design also shows to be superior to the systems available in the literature in several factors. Further studies need to be done to quantify the reliability of the system in clinical environments.

A dynamic 3D foot reconstruction system.

Foot problems are varied and range from simple disorders through to complex diseases and joint deformities. Wherever possible, the use of insoles, or orthoses, is preferred over surgery. Current insole design techniques are based on static measurements of the foot, despite the fact that orthoses are prevalently used in dynamic conditions while walking or running. This paper presents the design and implementation of a structured-light prototype system providing dense three dimensional (3D) measurements of the foot in motion, and its use to show that foot measurements in dynamic conditions differ significantly from their static counterparts. The input to the system is a video sequence of a foot during a single step; the output is a 3D reconstruction of the plantar surface of the foot for each frame of the input. Engineering and clinical tests were carried out for the validation of the system. The accuracy of the system was found to be 0.34 mm with planar test objects. In tests with real feet, the system proved repeatable, with reconstruction differences between trials one week apart averaging 2.44 mm (static case) and 2.81 mm (dynamic case). Furthermore, a study was performed to compare the effective length of the foot between static and dynamic reconstructions using the 4D system. Results showed an average increase of 9 mm for the dynamic case. This increase is substantial for orthotics design, cannot be captured by a static system, and its subject-specific measurement is crucial for the design of effective foot orthoses.