Research Highlights

Accurate and reliable tumor estimation using Deep Learning

AI and Deep Learning have been used to develop a solution for automated analysis and annotation of H&E tissue samples, identifying the boundary of the tumor and precisely measuring tissue cellularity and tumor cell content. The developed algorithms have now been expanded to automatically identify tumor in colorectal, melanoma, breast, and prostate tissue section. Trained on large datasets across multiple laboratories, the algorithms can drive automation of microdissection and quantitative analysis of percentage of tumor, providing an objective tissue quality evaluation for molecular pathology in solid tumors.

Related publications:

Serag A, et al. Translational AI and Deep Learning in Diagnostic Pathology. Front Med (Lausanne). 2019;6:185.

Anomaly Detection in digital pathology images using Generative Adversarial Networks (GANs)

Most deep learning methods require large annotated training datasets that are specific to a particular problem domain. Such large datasets are difficult to acquire for histopathology data where visual characteristics differ between different tissue types, besides the need for precise annotations. Serag et al. built an unsupervised learning to identify anomalies in histopathology imaging data as candidates for markers. The deep convolutional Generative Adversarial Network (GAN) learns a manifold of normal anatomical variability, accompanying an anomaly scoring scheme based on the mapping from image space to a latent space. Applied to new data, the model labels anomalies, and scores image patches indicating their fit into the learned distribution.

Related publications:

Serag A, Saint Martin M, Qureshi H, Diamond J, O'Reilly P, Hamilton P. 2019 September. Unsupervised anomaly detection: application to colorectal liver metastasis. In Virchows Archiv (Vol. 475, Pp. S59-S59). 233 Spring St, New York, NY 10013 USA: Springer.
Serag A, Ion-Margineanu A, Qureshi H, McMillan R, Saint Martin MJ, Diamond J, O'Reilly P, Hamilton P. Translational AI and Deep Learning in Diagnostic Pathology. Front Med (Lausanne). 2019;6:185.

Retinal Blood Vessels Extraction Using Probabilistic Modelling

The analysis of retinal blood vessels plays an important role in detecting and treating retinal diseases. In this review, we present an automated method to segment blood vessels of fundus retinal image. The proposed method could be used to support a non-intrusive diagnosis in modern ophthalmology for early detection of retinal diseases, treatment evaluation or clinical study. This study combines the bias correction and an adaptive histogram equalisation to enhance the appearance of the blood vessels. Then the blood vessels are extracted using probabilistic modelling that is optimised by the expectation maximisation algorithm. The method is evaluated on fundus retinal images of STARE and DRIVE datasets. The experimental results are compared with some recently published methods of retinal blood vessels segmentation. The experimental results show that our method achieved the best overall performance and it is comparable to the performance of human experts.

Related publications:

Kaba D, Wang C, Li Y, Salazar-Gonzalez A, Liu X, Serag A. Retinal blood vessels extraction using probabilistic modelling. Health Inf Sci Syst. 2014 Jan;2:2. https://doi.org/10.1186/2047-2501-2-2

4D modeling and analysis of brain development

Brain atlases are widely used in the neuroscience community as a tool for providing a standard space for comparison of subjects. However, adult brain atlases do not adequately represent the maturational patterns of the developing brain, and the use of an adult model in studying early brain growth may introduce substantial bias. Therefore, in the literature, several researchers have proposed to develop a digital atlas of the developing brain, still most efforts suffer from the comparatively lower level of anatomic definition and the coverage of a relatively narrow age range. Serag et al. developed an approach for constructing high-definition 4D atlases of the developing brain. A novelty in the approach is the use of a time-varying kernel width, to overcome the variations in the distribution of subjects at different ages. This leads to an atlas that retains a consistent level of detail at every time-point. Spatio-temporal brain atlases have been constructed for different groups using MR images from 284 subjects: 204 premature neonates between 28 and 44 weeks post-menstrual age at time of scan, and 80 fetuses between 23 and 37 weeks gestational age at time of scan. The resulting 4D fetal and neonatal atlases have greater anatomic definition than currently available 4D atlases, an important factor in improving registrations between the atlas and individual subjects with clear anatomical structures and atlas-based automatic segmentation. The fetal atlas provides a natural benchmark for assessing preterm born neonates and gives some insight into differences between the groups. Click here to download the atlases

Related publications:

Serag A, Aljabar P, Ball G, Counsell SJ, Boardman JP, Rutherford MA, Edwards AD, Hajnal JV, Rueckert D. Construction of a consistent high-definition spatio-temporal atlas of the developing brain using adaptive kernel regression. Neuroimage. 2012 Feb 1;59(3):2255-65. https://doi.org/10.1016/j.neuroimage.2011.09.062
Serag A, Kyriakopoulou V, Rutherford MA, Edwards AD, Hajnal JV, Aljabar P, Counsell SJ, Boardman JP, Rueckert D. A multi-channel 4D Probabilistic Atlas of the developing brain: Application to fetuses and neonates. Annals of BMVA. 2012;2012(3):1-14. http://www.bmva.org/annals/2012/2012-0003.pdf

GAMA: Gradients for Automated Motion correction and Analysis of the fetal brain

Quantitative volumes from brain magnetic resonance imaging (MRI) acquired across the life course may be useful for investigating long term effects of risk and resilience factors for brain development and healthy aging, and for understanding early life determinants of adult brain structure. Therefore, there is an increasing need for automated segmentation tools that can be applied to images acquired at different life stages. We developed an automatic segmentation method for human brain MRI, where a sliding window approach and a multi-class random forest classifier were applied to high-dimensional feature vectors for accurate segmentation. The method performed well on brain MRI data acquired from 179 individuals, analyzed in three age groups: newborns (38–42 weeks gestational age), children and adolescents (4–17 years) and adults (35–71 years). As the method can learn from partially labeled datasets, it can be used to segment large-scale datasets efficiently. It could also be applied to different populations and imaging modalities across the life course.

Related publications:

Serag A, Macnaught G, Denison FC, Reynolds RM, Semple SI, Boardman JP. Histograms of Oriented 3D Gradients for Fully Automated Fetal Brain Localization and Robust Motion Correction in 3 T Magnetic Resonance Images. Biomed Res Int. 2017;2017:3956363. https://doi.org/10.1155/2017/3956363

SEGMA: An Automatic SEGMentation Approach for Human Brain MRI

Serag A, Wilkinson AG, Telford EJ, Pataky R, Sparrow SA, Anblagan D, Macnaught G, Semple SI, Boardman JP. SEGMA: An Automatic SEGMentation Approach for Human Brain MRI using Sliding Window and Random Forests. Front Neuroinform. 2017;11:2. https://doi.org/10.3389/fninf.2017.00002

ALFA: Accurate Learning with Few Atlases

Accurate whole-brain segmentation, or brain extraction, of magnetic resonance imaging (MRI) is a critical first step in most neuroimage analysis pipelines. The majority of brain extraction algorithms have been developed and evaluated for adult data and their validity for neonatal brain extraction, which presents age-specific challenges for this task, has not been established. We developed a novel method for brain extraction of multi-modal neonatal brain MR images, named ALFA (Accurate Learning with Few Atlases). The method uses a new sparsity-based atlas selection strategy that requires a very limited number of atlases ‘uniformly’ distributed in the low-dimensional data space, combined with a machine learning based label fusion technique. The performance of the method for brain extraction from multi-modal data of 50 newborns is evaluated and compared with results obtained using eleven publicly available brain extraction methods. ALFA outperformed the eleven compared methods providing robust and accurate brain extraction results across different modalities. As ALFA can learn from partially labelled datasets, it can be used to segment large-scale datasets efficiently. ALFA could also be applied to other imaging modalities and other stages across the life course.

Related publications:

Serag A, Blesa M, Moore EJ, Pataky R, Sparrow SA, Wilkinson AG, Macnaught G, Semple SI, Boardman JP. Accurate Learning with Few Atlases (ALFA): An algorithm for MRI neonatal brain extraction and comparison with 11 publicly available methods. Sci Rep. 2016 Mar 24;6:23470. https://doi.org/10.1038/srep23470

Longitudinal image registration via spatio-temporal atlase

A novel framework is proposed for longitudinal registration which can handle large intra-subject anatomical variations. The framework exploits freely available spatio-temporal atlases, which can aid the longitudinal registration process as it provides prior information about the missing anatomical evolution between two scans taken over large time-interval. The spatio-temporal atlas is used to develop an approach to carry out longitudinal registrations via atlas propagation. Evaluation experiments have been carried out using 50 neonatal subjects with two scans taken at different time-intervals. Different metrics were used, including intensity similarity and segmentation overlap, to assess the registration performance. The proposed registration framework outperforms direct registration by providing an accurate and consistent registration, particularly when the time-interval between scans increases. The approach is refered to as a framework for Longitudinal Image registration via Spatio-temporal Atlases (LISA).

Related publications:

Serag A, Aljabar P, Counsell S, Boardman J, Hajnal JV, Rueckert D. LISA: Longitudinal image registration via spatio-temporal atlases. 2012 9th IEEE International Symposium on Biomedical Imaging (ISBI). 2012 May:334-7. https://doi.org/10.1109/ISBI.2012.6235552

Modeling the appearance variation of the perinatal brain

Preterm birth is associated with abnormal brain development and long-term neurodevelopmental impairment. Quantitative magnetic resonance (MR) studies of preterm brain injury have focused on morphological features such as shape and volume and on measures of tissue microstructure obtained from diffusion tensor imaging. This work focuses on longitudinal changes in signal intensity, which can offer a useful marker for mapping developmental changes. The proposed analysis framework utilises spatial normalization, intensity normalization and kernel regression. Changes over time in T1- and T2-weighted signal intensity were measured in subcortical grey matter and white matter. The study shows that quantitative signal change analysis on a large cohort is feasible, and that it can serve as a marker for developmental brain changes, both normal and abnormal, which might ultimately lead to a better understanding of the trajectory of early brain maturation.

Related publications:

Serag A, Aljabar P, Ball G, Counsell SJ, Boardman JP, Hajnal JV, Rueckert D. Developmental signal intensity changes in subcortical structures of the perinatal brain detected using Multi-modal MRI. MICCAI workshop on Spatiotemporal Image Analysis for Longitudinal and Time-Series Image Data - STIA. 2010.

Unsupervised learning of shape complexity: Application to Brain Development

A framework is presented for unsupervised learning of shape complexity in the developing brain. It learns the complexity in different brain structures by applying several shape complexity measures to each individual structure, and then using feature selection to select the measures that best describe the changes in complexity of each structure. Then, feature selection is applied again to assign a score to each structure, in order to find which structure can be a good biomarker of brain development. This study was carried out using T2-weighted MR images from 224 premature neonates (the age range at the time of scan was 26.7 to 44.86 weeks post-menstrual age). The advantage of the proposed framework is that one can extract as many ROIs as desired, and the framework automatically finds the ones which can be used as good biomarkers. However, the example application focuses on neonatal brain image data, the proposed framework for combining information from multiple measures may be applied more generally to other populations and other forms of imaging data.

Related publications:

Serag A, Gousias IS, Makropoulos A, Aljabar P, Hajnal JV, Boardman JP, Counsell SJ, Rueckert D. Unsupervised learning of shape complexity: Application to brain development. International Workshop on Spatio-temporal Image Analysis for Longitudinal and Time-Series Image Data. 2012;7570(Second International Workshop, STIA 2012):88-99. https://doi.org/10.1007/978-3-642- 33555-6_8

Optimized methodology for neonatal diffusion tensor imaging processing and study-specific template construction

Diffusion tensor imaging (DTI) has been widely used to study cerebral white matter microstructure in vivo. There is a plethora of open source tools available to perform pre-processing, analysis and template or atlas construction, however very few have been optimized for use with neonatal DTI data. Here we present a fully automated modular pipeline optimized for neonatal DTI data and the construction of study-specific tensor templates. We compare our methodology to an existing one. It is anticipated that the construction of population or study-specific templates will facilitate better group comparisons of neonatal populations both in health and disease.

Related publications:

Evangelou IE, Serag A, Bouyssi-Kobar M, Plessis AJ, Limperopoulos C. Optimized methodology for neonatal diffusion tensor imaging processing and study-specific template construction. Annu Int Conf IEEE Eng Med Biol Soc. 2014;2014:2372-5. https://doi.org/10.1109/embc.2014.6944098

Robust motion correction and outlier rejection of in vivo functional MR images of the fetal brain and placenta during maternal hyperoxia

Subject motion is a major challenge in functional magnetic resonance imaging studies (fMRI) of the fetal brain and placenta during maternal hyperoxia. We propose a motion correction and volume outlier rejection method for the correction of severe motion artifacts in both fetal brain and placenta. The method is optimized to the experimental design by processing different phases of acquisition separately. It also automatically excludes high-motion volumes and all the missing data are regressed from ROI-averaged signals. The results demonstrate that the proposed method is effective in enhancing motion correction in fetal fMRI without large data loss, compared to traditional motion correction methods.

Related publications:

You W, Serag A, Evangelou I, Andescavage N, Limperopoulos C. Robust motion correction and outlier rejection of in vivo functional MR images of the fetal brain and placenta during maternal hyperoxia. SPIE Medical Imaging. 2015 May;9417(Medical Imaging 2015: Biomedical Applications in Molecular, Structural, and Functional Imaging). https://doi.org/10.1117/12.2082451

Parcellation of neonatal brain MRI into 107 regions using atlas propagation through intermediate time points in childhood

Neuroimaging analysis pipelines rely on parcellated atlases generated from healthy individuals to provide anatomical context to structural and diffusion MRI data (dMRI). Longitudinal studies of brain development require an atlas that is operable under conditions of wide anatomical variation. We present the Edinburgh Neonatal Atlas (ENA25), an MRI neonatal brain atlas created from 25 healthy subjects using temporal registration of an adult atlas through intermediate time points from 4.5 years to the neonatal period. The atlas contains 107 cortical and subcortical regions, which provides the greatest anatomical detail produced to our knowledge, and we provide normative volumetric, dMRI metrics and grey matter / white matter probabilities for these regions. Templates for different MRI modalities (structural and diffusion MRI) and symmetric templates, which are required for studies of laterality are provided.

Related publications:

Blesa M, Serag A, Wilkinson AG, Anblagan D, Telford EJ, Pataky R, Sparrow SA, Macnaught G, Semple SI, Bastin ME, Boardman JP. Parcellation of the healthy neonatal brain into 107 regions using Atlas Propagation through Intermediate Time Points in childhood. Front Neurosci. 2016 Jun;10:220. https://doi.org/10.3389/fnins.2016.00220