Deep learning for Alzheimer's disease: Mapping large-scale histological tau protein for neuroimaging biomarker validation.

Abnormal tau inclusions are hallmarks of Alzheimer's disease and predictors of clinical decline. Several tau PET tracers are available for neurodegenerative disease research, opening avenues for molecular diagnosis in vivo. However, few have been approved for clinical use. Understanding the neurobiological basis of PET signal validation remains problematic because it requires a large-scale, voxel-to-voxel correlation between PET and (immuno) histological signals. Large dimensionality of whole human brains, tissue deformation impacting co-registration, and computing requirements to process terabytes of information preclude proper validation. We developed a computational pipeline to identify and segment particles of interest in billion-pixel digital pathology images to generate quantitative, 3D density maps. The proposed convolutional neural network for immunohistochemistry samples, IHCNet, is at the pipeline's core. We have successfully processed and immunostained over 500 slides from two whole human brains with three phospho-tau antibodies (AT100, AT8, and MC1), spanning several terabytes of images. Our artificial neural network estimated tau inclusion from brain images, which performs with ROC AUC of 0.87, 0.85, and 0.91 for AT100, AT8, and MC1, respectively. Introspection studies further assessed the ability of our trained model to learn tau-related features. We present an end-to-end pipeline to create terabytes-large 3D tau inclusion density maps co-registered to MRI as a means to facilitate validation of PET tracers.

High thickness histological sections as alternative to study the three-dimensional microscopic human sub-cortical neuroanatomy.

Stereotaxy is based on the precise image-guided spatial localization of targets within the human brain. Even with the recent advances in MRI technology, histological examination renders different (and complementary) information of the nervous tissue. Although several maps have been selected as a basis for correlating imaging results with the anatomical locations of sub-cortical structures, technical limitations interfere in a point-to-point correlation between imaging and anatomy due to the lack of precise correction for post-mortem tissue deformations caused by tissue fixation and processing. We present an alternative method to parcellate human brain cytoarchitectural regions, minimizing deformations caused by post-mortem and tissue-processing artifacts and enhancing segmentation by means of modified high thickness histological techniques and registration with MRI of the same specimen and into MNI space (ICBM152). A three-dimensional (3D) histological atlas of the human thalamus, basal ganglia, and basal forebrain cholinergic system is displayed. Structure's segmentations were performed in high-resolution dark-field and light-field microscopy. Bidimensional non-linear registration of the histological slices was followed by 3D registration with in situ MRI of the same subject. Manual and automated registration procedures were adopted and compared. To evaluate the quality of the registration procedures, Dice similarity coefficient and normalized weighted spectral distance were calculated and the results indicate good overlap between registered volumes and a small shape difference between them in both manual and automated registration methods. High thickness high-resolution histological slices in combination with registration to in situ MRI of the same subject provide an effective alternative method to study nuclear boundaries in the human brain, enhancing segmentation and demanding less resources and time for tissue processing than traditional methods.

Automating cell detection and classification in human brain fluorescent microscopy images using dictionary learning and sparse coding.

BACKGROUND: Immunofluorescence (IF) plays a major role in quantifying protein expression in situ and understanding cell function. It is widely applied in assessing disease mechanisms and in drug discovery research. Automation of IF analysis can transform studies using experimental cell models. However, IF analysis of postmortem human tissue relies mostly on manual interaction, often subjected to low-throughput and prone to error, leading to low inter and intra-observer reproducibility. Human postmortem brain samples challenges neuroscientists because of the high level of autofluorescence caused by accumulation of lipofuscin pigment during aging, hindering systematic analyses. We propose a method for automating cell counting and classification in IF microscopy of human postmortem brains. Our algorithm speeds up the quantification task while improving reproducibility. NEW METHOD: Dictionary learning and sparse coding allow for constructing improved cell representations using IF images. These models are input for detection and segmentation methods. Classification occurs by means of color distances between cells and a learned set. RESULTS: Our method successfully detected and classified cells in 49 human brain images. We evaluated our results regarding true positive, false positive, false negative, precision, recall, false positive rate and F1 score metrics. We also measured user-experience and time saved compared to manual countings. COMPARISON WITH EXISTING METHODS: We compared our results to four open-access IF-based cell-counting tools available in the literature. Our method showed improved accuracy for all data samples. CONCLUSION: The proposed method satisfactorily detects and classifies cells from human postmortem brain IF images, with potential to be generalized for applications in other counting tasks.

Computerized brain tumor segmentation in magnetic resonance imaging.

OBJECTIVE: To propose an automatic brain tumor segmentation system. METHODS: The system used texture characteristics as its main source of information for segmentation. RESULTS: The mean correct match was 94% of correspondence between the segmented areas and ground truth. CONCLUSION: Final results showed that the proposed system was able to find and delimit tumor areas without requiring any user interaction.

Computerized brain tumor segmentation in magnetic resonance imaging / Segmentação computadorizada de tumores do encéfalo em imagens de ressonância magnética

Objective: To propose an automatic brain tumor segmentation system. Methods: The system used texture characteristics as its main source of information for segmentation. Results: The mean correct match was 94% of correspondence between the segmented areas and ground truth. Conclusion: Final results showed that the proposed system was able to find and delimit tumor areas without requiring any user interaction.

Objetivo: Propor um sistema para segmentação automática de tumores do encéfalo. Métodos: O sistema emprega parâmetros de textura como sua principal fonte de informação para a segmentação. Resultados: Os acertos chegaram a 94% na correspondência entre a segmentação obtida e o padrão-ouro. Conclusão: Os resultados obtidos mostram que o sistema é capaz de localizar e delimitar as áreas de tumor sem necessidade de interação com o operador.

Texture analysis of high resolution MRI allows discrimination between febrile and afebrile initial precipitating injury in mesial temporal sclerosis.

A computational pipeline combining texture analysis and pattern classification algorithms was developed for investigating associations between high-resolution MRI features and histological data. This methodology was tested in the study of dentate gyrus images of sclerotic hippocampi resected from refractory epilepsy patients. Images were acquired using a simple surface coil in a 3.0T MRI scanner. All specimens were subsequently submitted to histological semiquantitative evaluation. The computational pipeline was applied for classifying pixels according to: a) dentate gyrus histological parameters and b) patients' febrile or afebrile initial precipitating insult history. The pipeline results for febrile and afebrile patients achieved 70% classification accuracy, with 78% sensitivity and 80% specificity [area under the reader observer characteristics (ROC) curve: 0.89]. The analysis of the histological data alone was not sufficient to achieve significant power to separate febrile and afebrile groups. Interesting enough, the results from our approach did not show significant correlation with histological parameters (which per se were not enough to classify patient groups). These results showed the potential of adding computational texture analysis together with classification methods for detecting subtle MRI signal differences, a method sufficient to provide good clinical classification. A wide range of applications of this pipeline can also be used in other areas of medical imaging.

Hippocampal CA3 transcriptome signature correlates with initial precipitating injury in refractory mesial temporal lobe epilepsy.

BACKGROUND: Prolonged febrile seizures constitute an initial precipitating injury (IPI) commonly associated with refractory mesial temporal lobe epilepsy (RMTLE). In order to investigate IPI influence on the transcriptional phenotype underlying RMTLE we comparatively analyzed the transcriptomic signatures of CA3 explants surgically obtained from RMTLE patients with (FS) or without (NFS) febrile seizure history. Texture analyses on MRI images of dentate gyrus were conducted in a subset of surgically removed sclerotic hippocampi for identifying IPI-associated histo-radiological alterations. METHODOLOGY/PRINCIPAL FINDINGS: DNA microarray analysis revealed that CA3 global gene expression differed significantly between FS and NFS subgroups. An integrative functional genomics methodology was used for characterizing the relations between GO biological processes themes and constructing transcriptional interaction networks defining the FS and NFS transcriptomic signatures and its major gene-gene links (hubs). Co-expression network analysis showed that: i) CA3 transcriptomic profiles differ according to the IPI; ii) FS distinctive hubs are mostly linked to glutamatergic signalization while NFS hubs predominantly involve GABAergic pathways and neurotransmission modulation. Both networks have relevant hubs related to nervous system development, what is consistent with cell genesis activity in the hippocampus of RMTLE patients. Moreover, two candidate genes for therapeutic targeting came out from this analysis: SSTR1, a relevant common hub in febrile and afebrile transcriptomes, and CHRM3, due to its putative role in epilepsy susceptibility development. MRI texture analysis allowed an overall accuracy of 90% for pixels correctly classified as belonging to FS or NFS groups. Histological examination revealed that granule cell loss was significantly higher in FS hippocampi. CONCLUSIONS/SIGNIFICANCE: CA3 transcriptional signatures and dentate gyrus morphology fairly correlate with IPI in RMTLE, indicating that FS-RMTLE represents a distinct phenotype. These findings may shed light on the molecular mechanisms underlying refractory epilepsy phenotypes and contribute to the discovery of novel specific drug targets for therapeutic interventions.