Research Assistant - computer vision

Radiology Department at the College of Medicine and Health Sciences of the United Arab Emirates University (UAEU) is announcing the position of a research assistant in clinical research and, in particular, medical imaging. The research is to perform diagnostic and predictive models of various diseases by applying conventional and deep learning approaches to demographic, environmental and multimodal medical data. Tasks to accomplish: 1. Preprocessing of medical images: de-identifying the images, data preparation and denoising of the diagnostics images, 2D and 3D data scaling and normalization, the development of image data loaders in PyTorch or its analogs, the preprocessing of numerical and textual data in the images, replacing/removing missing values; testing the data for outliers, finding duplicated or highly correlated features; applying masks of the studied organ to the diagnostic images, 2. Running classification, detection, and segmentation tasks (with 3DSlicer, CAT12, etc.). 3. Preparing analytics over the preprocessed dataset: checking whether the dataset is versatile enough for different ML approaches; finding correlations between different features; running feature selection algorithms to find the most informative data sources (numerical vs. images) 4. Building up regression and classification models and estimating their performance metrics: preparing a data visualization module for the qualitative and quantitative quality validation, training a deep convolutional neural net to get predictions based on 2D and 3D data, training models with spatial attention (standard and self-attention) to focus on the most valuable parts of 2D and 3D data, trying classical non-neural approaches (e.g. gradient boosting) to get predictions based on the whole dataset, adding functions for calculating ROC-AUC, error rate, precision, recall, sensitivity, specificity. 5. Applying deep learning-based architectures to analyze visual data in different variant: convolutional NN for classification tasks, single stage or multistage DLL detection systems, generative adversarial networks (GAN) for super-resolution and reconstruction, recurrent neural networks for automatic report generation.

MS/PhD degree in relevant sphere of studies will be of advantage

5000 - 6000 based on Qualification and Experience

before

the closing date.