A5016385771- Cell Image Analysis Techniques
- Radiomics and Machine Learning in Medical Imaging
- AI in cancer detection
- Advanced Neuroimaging Techniques and Applications
- Neuroinflammation and Neurodegeneration Mechanisms
- Machine Learning in Healthcare
- Generative Adversarial Networks and Image Synthesis
- Image and Signal Denoising Methods
- Explainable Artificial Intelligence (XAI)
- Advanced Image Processing Techniques
- Axon Guidance and Neuronal Signaling
- Domain Adaptation and Few-Shot Learning
- Brain Tumor Detection and Classification
- Cancer Genomics and Diagnostics
- Optical Imaging and Spectroscopy Techniques
- Nerve injury and regeneration
- Computational Drug Discovery Methods
- 3D Shape Modeling and Analysis
- Biomedical Text Mining and Ontologies
- Lung Cancer Diagnosis and Treatment
- Anesthesia and Neurotoxicity Research
- Neurogenesis and neuroplasticity mechanisms
- Image Processing Techniques and Applications
- Functional Brain Connectivity Studies
- Advanced Electron Microscopy Techniques and Applications
HiETA Technologies (United Kingdom)
2024-2025
Imperial College London
2017-2024
King's College London
2020-2022
MRC Unit for Lifelong Health and Ageing
2022
University College London
2022
King's College School
2022
Hammersmith Hospital
2017
Abstract Background The objective of this comprehensive pan-cancer study is to evaluate the potential deep learning (DL) for molecular profiling multi-omic biomarkers directly from hematoxylin and eosin (H&E)-stained whole slide images. Methods A total 12,093 DL models predicting 4031 across 32 cancer types were trained validated. included a broad range genetic, transcriptomic, proteomic biomarkers, as well established prognostic markers, subtypes, clinical outcomes. Results Here we show...
44 Background: Testing for microsatellite instability (MSI) or mismatch repair deficiency (dMMR) is part of the diagnosis and clinical management patients with colorectal cancer (CRC). Healthcare services recommend MSI dMMR testing all CRC to guide therapeutic choices assist in identifying Lynch Syndrome. However, practice, high costs demand timely test results, combined rising prevalence a shrinking pathology workforce, present barrier universal adoption. This highlights need rapid...
An important goal of medical imaging is to be able precisely detect patterns disease specific individual scans; however, this challenged in brain by the degree heterogeneity shape and appearance. Traditional methods, based on image registration, historically fail variable features disease, as they utilise population-based analyses, suited primarily studying group-average effects. In paper we therefore take advantage recent developments generative deep learning develop a method for...
Abstract The emerging field of geometric deep learning extends the application convolutional neural networks to irregular domains such as graphs, meshes and surfaces. Several recent studies have explored potential for using these techniques analyse segment cortical surface. However, there has been no comprehensive comparison approaches one another, nor existing Euclidean methods, date. This paper benchmarks a collection traditional models on phenotype prediction segmentation sphericalised...
Studies of structural plasticity in the brain often require detection and analysis axonal synapses (boutons). To date, bouton has been largely manual or semi-automated, relying on a step that traces axons before boutons. If tracing axon fails, accuracy is compromised. In this paper, we propose new algorithm does not to detect boutons 3D two-photon images taken from mouse cortex. find most appropriate techniques for task, compared several well-known algorithms interest point feature...
Feature attribution (FA), or the assignment of class-relevance to different locations in an image, is important for many classification problems but particularly crucial within neuroscience domain, where accurate mechanistic models behaviours, disease, require knowledge all features discriminative a trait. At same time, predicting class relevance from brain images challenging as phenotypes are typically heterogeneous, and changes occur against background significant natural variation. Here,...
Abstract We assessed the pan-cancer predictability of multi-omic biomarkers from haematoxylin and eosin (H&E)-stained whole slide images (WSI) using deep learning (DL) throughout a systematic study. A total 13,443 DL models predicting 4,481 across 32 cancer types were trained validated. The investigated included broad range genetic, transcriptomic, proteomic, metabolic alterations, as well established markers relevant for prognosis, molecular subtypes clinical outcomes. Overall, we found...
Abstract We present a public validation of PANProfiler (ER, PR, HER2), an in-vitro medical device (IVD) that predicts the qualitative status estrogen receptor (ER), progesterone (PR) and human epidermal growth factor 2 (HER2) by analysing hematoxylin eosin (H&E)-stained tissue scan. In on 648 560 unseen cases with known biomarker status, achieves accuracy 87% 83% (HER2). The offers early evidence ability to predict clinically relevant breast biomarkers from H&E slide in clinical setting.
Abstract Background Despite the widespread occurrence of axon and synaptic loss in injured diseased nervous system, cellular molecular mechanisms these key degenerative processes remain incompletely understood. Wallerian degeneration (WD) is a tightly regulated form after injury, which has been intensively studied large myelinated fibre tracts spinal cord, optic nerve peripheral system (PNS). Fewer studies, however, have focused on WD complex neuronal circuits mammalian brain, were mainly...
Recent developments in deep networks allow us to train with more parameters by yielding better performance given sufficient amount of data. However, we are still restricted the availability labelled data medical image segmentation, where problem is exacerbated high intra- and intervariability anatomical structures. In order bypass this without compromising network performance, study introduces a PERINet, which promises achieve higher while being smaller parameter count such as on 0.8 million...
We present a proof-of-concept, deep learning (DL) based, differentiable biomechanical model of realistic brain deformations. Using prescribed maps local atrophy and growth as input, the network learns to deform images according Neo-Hookean tissue deformation. The tool is validated using longitudinal data from Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset, we demonstrate that trained capable rapidly simulating new deformations with minimal residuals. This method has potential be...
How aging affects axon regeneration and synaptic repair in the brain is poorly understood. To study age-related changes neural circuits, we developed a model of axonal injury aged (> 2 years) mouse somatosensory cortex. By directly tracking fluorescently labelled injured axons by multiphoton imaging, find that while degeneration comparable to young adult brain, impaired. We further examine most common type cortical synapses, En Passant Boutons (EPBs), observe transient significant...
562 Background: Newly diagnosed breast cancer specimens are routinely tested for oestrogen receptor (ER), progesterone (PR) and human epidermal growth factor receptor-2 (HER2) status. This requires additional immunohistochemistry (IHC) +/- in situ hybridisation (ISH) assessment. can increase laboratory/pathologist workloads turnaround times. A computer-based approach could help to alleviate these issues. PANProfiler Breast is a deep-learning solution designed predict ER/PR status detect HER2...
ABSTRACT Despite the widespread occurrence of axon degeneration in injured and diseased nervous system, mechanisms degenerative process remain incompletely understood. In particular, factors that regulate how individual axons degenerate within their native environment mammalian brain are unknown. Longitudinal imaging >120 individually cortical revealed a threshold length below which undergo rapid-onset form Wallerian (ROWD). ROWD consistently starts 10 times earlier is executed 4 slower...