Deep Learning & Image Analysis
An image is worth 1,000 datapoints
Deep Learning is a powerful machine learning technology that can extract knowledge and create predictive models using images as input information. Based on this technology, the NanoSolveIT consortium has developed computational workflows for the prediction of adverse effects due to exposure to ENMs, which are offered through the platform along with other tools for image analysis and feature extraction.
Available Services
Deep Learning model to predict NM exposure effects on Daphnia magna
Deep learning models for predicting the effects of exposure to engineered nanomaterials on Daphnia magna. A workflow applying two different deep learning architectures on microscopic images of Daphnia magna is developed that can automatically detect the length of the tail, and the overall size, and uncommon lipid concentrations and lipid deposit shapes on the heart and the abdomen / claw of the Daphnia magna.
NanoXtract: a unique online tool for the calculation of image descriptors based on TEM images of NMs
Analyse TEM NMs images to extract useful descriptors that can be used to provide a much more nuanced description of the sample composition and/or as inputs to toxicity prediction models.
Sbpot: Predicting the median (%) strand breaks, estimated from built-in microscope image analysis tools
AdaBoost model (using an Extra-Trees Regressor as a base estimator) to predict toxicity due to exposure to NMs. Developed on data collected from Comet Assay images, the model can make predictions on the toxicity of NMs using as a response variable the median (%) strand breaks, estimated from built-in microscope image analysis tools. High values of the response variable suggest increased NM toxicity.
NanoImage: An automated tool for extracting descriptors from electronic images
The NanoImage tool processes microscopy nanomaterial images of Spherical Particles, Nanotubes and images with a high occurrence of Aggregates/Agglomerates. As both the nanomaterials and the resulting microscopy images can vary greatly, users supply certain parameters to adjust the output to their preference and scope of analysis: these comprise the choice of filter to be applied and the selection of descriptors to be calculated. The application outputs a table with the calculated descriptor values for each particle and the average value for all particles of the image, which can be produced in a range of transferable formats.
DeepSegm: Deep learning models automating the segmentation of agglomerated, non-spherical particles from electron microscopy images
DeepSegm is a web implementation of the deep learning model, which has been presented in the publication: Rühle, B., Krumrey, J.F. & Hodoroaba, VD. Sci Rep 11, 4942 (2021). The tool automates the process of segmentation of agglomerated, non-spherical particles from electron microscopy images. The image segmentation process is based on Generative Adversarial Networks (GAN) and MultiRes Unet deep learning methods technologies.