A5085313426- Generative Adversarial Networks and Image Synthesis
- Model Reduction and Neural Networks
- Radiomics and Machine Learning in Medical Imaging
- Adversarial Robustness in Machine Learning
- Domain Adaptation and Few-Shot Learning
- Medical Imaging Techniques and Applications
- Machine Learning in Healthcare
- Advanced Neural Network Applications
- Advanced X-ray Imaging Techniques
- Machine Learning in Materials Science
- Computational Physics and Python Applications
- Anomaly Detection Techniques and Applications
- Numerical methods in inverse problems
- Ovarian cancer diagnosis and treatment
- Image and Signal Denoising Methods
- COVID-19 diagnosis using AI
- Algorithms and Data Compression
- Artificial Intelligence in Healthcare and Education
- Metabolomics and Mass Spectrometry Studies
- Advanced Vision and Imaging
- MRI in cancer diagnosis
- Ultrasound Imaging and Elastography
- Medical Imaging and Analysis
- Explainable Artificial Intelligence (XAI)
- Computational Drug Discovery Methods
University of Cambridge
2020-2024
University of Bath
2020
University of Bremen
2017-2019
Machine learning methods offer great promise for fast and accurate detection prognostication of COVID-19 from standard-of-care chest radiographs (CXR) computed tomography (CT) images. Many articles have been published in 2020 describing new machine learning-based models both these tasks, but it is unclear which are potential clinical utility. In this systematic review, we search EMBASE via OVID, MEDLINE PubMed, bioRxiv, medRxiv arXiv papers preprints uploaded January 1, to October 3,...
Abstract Motivation Tumor classification using imaging mass spectrometry (IMS) data has a high potential for future applications in pathology. Due to the complexity and size of data, automated feature extraction steps are required fully process data. Since spectra exhibit certain structural similarities image deep learning may offer promising strategy IMS as it been successfully applied classification. Results Methodologically, we propose an adapted architecture based on convolutional...
Recent studies on the adversarial vulnerability of neural networks have shown that models trained to be more robust attacks exhibit interpretable saliency maps than their non-robust counterparts. We aim quantify this behavior by considering alignment between input image and map. hypothesize as distance decision boundary grows,so does alignment. This connection is strictly true in case linear models. confirm these theoretical findings with experiments based a local Lipschitz regularization...
Score-based diffusion models have emerged as one of the most promising frameworks for deep generative modelling. In this work we conduct a systematic comparison and theoretical analysis different approaches to learning conditional probability distributions with score-based models. particular, prove results which provide justification successful estimators score. Moreover, introduce multi-speed framework, leads new estimator score, performing on par previous state-of-the-art approaches. Our...
Over the past few years, deep learning has risen to foreground as a topic of massive interest, mainly result successes obtained in solving large-scale image processing tasks. There are multiple challenging mathematical problems involved applying learning: most methods require solution hard optimisation problems, and good understanding trade-off between computational effort, amount data model complexity is required successfully design approach for given problem.. A large progress made been...
U-Nets have been established as a standard neural network architecture for image-to-image problems such segmentation and inverse in imaging. For high-dimensional applications, they example appear 3D medical imaging, however prohibitive memory requirements. Here, we present new fully-invertible U-Net-based called the iUNet, which allows application of highly memory-efficient backpropagation procedures. As its main building block, introduce learnable invertible up- an downsampling operations....
In recent years the use of convolutional layers to encode an inductive bias (translational equivariance) in neural networks has proven be a very fruitful idea. The successes this approach have motivated line research into incorporating other symmetries deep learning methods, form group equivariant networks. Much work been focused on roto-translational symmetry $\mathbf R^d$, but examples are scaling R^d$ and rotational sphere. work, we demonstrate that operations can naturally incorporated...
Wasserstein GANs are based on the idea of minimising distance between a real and generated distribution. We provide an in-depth mathematical analysis differences theoretical setup reality training GANs. In this work, we gather both empirical evidence that WGAN loss is not meaningful approximation distance. Moreover, argue even desirable function for deep generative models, conclude success can in truth be attributed to failure approximate
Abstract Purpose To determine if pelvic/ovarian and omental lesions of ovarian cancer can be reliably segmented on computed tomography (CT) using fully automated deep learning-based methods. Methods A learning model for the two most common disease sites high-grade serous (pelvis/ovaries omentum) was developed compared against well-established “no-new-Net” framework unrevised trainee radiologist segmentations. total 451 CT scans collected from four different institutions were used training (...
U-Nets have been established as a standard architecture for image-to-image learning problems such segmentation and inverse in imaging. For large-scale data, it example appears 3D medical imaging, the U-Net however has prohibitive memory requirements. Here, we present new fully-invertible U-Net-based called iUNet, which employs novel learnable invertible up- downsampling operations, thereby making use of memory-efficient backpropagation possible. This allows us to train deeper larger networks...
Abstract Purpose To determine if pelvic/ovarian and omental lesions of ovarian cancer can be reliably segmented on computed tomography (CT) using fully automated deep learning-based methods. Materials Methods A learning model for the two most common disease sites high grade serous (pelvis/ovaries omentum) was developed compared against well-established “no-new-Net” (nnU-Net) framework unrevised trainee radiologist segmentations. total 451 pre-treatment post neoadjuvant chemotherapy (NACT) CT...
Diffusion models have emerged as one of the most promising frameworks for deep generative modeling. In this work, we explore potential non-uniform diffusion models. We show that leads to multi-scale which similar structure normalizing flows. experimentally find in same or less training time, model achieves better FID score than standard uniform model. More importantly, it generates samples $4.4$ times faster $128\times 128$ resolution. The speed-up is expected be higher resolutions where...
In recent years, an increasing number of neural network models have included derivatives with respect to inputs in their loss functions, resulting so-called double backpropagation for first-order optimization. However, so far no general description the involved exists. Here, we cover a wide array special cases very Hilbert space framework, which allows us provide optimized rules many real-world scenarios. This includes reduction calculations Frobenius-norm-penalties on Jacobians by roughly...
Abstract Purpose Breast cancer is the most common malignancy in women. Unfortunately, current breast imaging techniques all suffer from certain limitations: they are either not fully three dimensional, have an insufficient resolution or low soft‐tissue contrast. Grating interferometry computed tomography (GI‐BCT) a promising X‐ray phase contrast modality that could overcome these limitations by offering high and excellent three‐dimensional resolution. To enable transition of this technology...
We introduce CAFLOW, a new diverse image-to-image translation model that simultaneously leverages the power of auto-regressive modeling and efficiency conditional normalizing flows. transform conditioning image into sequence latent encodings using multi-scale flow repeat process for conditioned image. distribution by distributions with an efficient flow, where each factor affects synthesis at its respective resolution scale. Our proposed framework performs well on range tasks. It outperforms...
We introduce CAFLOW, a new diverse image-to-image translation model that simultaneously leverages the power of auto-regressive modeling and efficiency conditional normalizing flows. transform conditioning image into sequence latent encodings using multi-scale flow repeat process for conditioned image. distribution by distributions with an efficient flow, where each factor affects synthesis at its respective resolution scale. Our proposed framework performs well on range tasks. It outperforms...
An increasing number of models require the control spectral norm convolutional layers a neural network. While there is an abundance methods for estimating and enforcing upper bounds on those during training, they are typically costly in either memory or time. In this work, we introduce very simple method normalization depthwise separable convolutions, which introduces negligible computational overhead. We demonstrate effectiveness our image classification tasks using standard architectures...
Neural networks have recently been established as a viable classification method for imaging mass spectrometry data tumor typing. For multi-laboratory scenarios however, certain confounding factors may strongly impede their performance. In this work, we introduce Deep Relevance Regularization, of restricting what the neural network can focus on during classification, in order to improve We demonstrate how Regularization robustifies against challenging inter-lab dataset consisting breast and...
Over the past few years, deep learning has risen to foreground as a topic of massive interest, mainly result successes obtained in solving large-scale image processing tasks. There are multiple challenging mathematical problems involved applying learning: most methods require solution hard optimisation problems, and good understanding tradeoff between computational effort, amount data model complexity is required successfully design approach for given problem. A large progress made been...