Search

We propose a novel multiple-input multiple-output convolution neural network (MIMO-Net) for cell segmentation in fluorescence microscopy images. The proposed network trains the network parameters using multiple resolutions of the input image, connects the intermediate layers for better localization and context and generates the output using multi-resolution deconvolution filters. The MIMO-Net allows us to deal with variable intensity cell boundaries and highly variable cell size in the mouse pancreatic tissue by adding extra convolutional layers which bypass the max-pooling operation. The results show that our method outperforms state-of-the-art deep learning based approaches for segmentation.

The application of multi-tag protocols in fluorescence microscopy allows the visualization of a large number (> 10) of molecules (i. e. proteins) in a sample (like a tissue section). However, the analysis of such high dimensional bioimages is a difficult task for most of the labs, since software solutions for particular data mining steps are difficult to use or just not available. In this paper we present two new free online tools: MICOLT (Multivariate Image COlocation Tool) and MIFIST (Multivariate Image Frequent Item Set Tool). Both tools can be used via our recently proposed online bioimage analysis platform BioIMAX, so users can upload their bioimage data, apply the tools and share the results with other invited users based on BioIMAX' concept of shared virtual projects. Data mining with these tools includes the computation and visualization colocation factors well established in the microscopy community (like Mander's score) and association rule mining following the frequent item set principle, thereby supporting large and small scale analysis.

Fluorescence microscopy has regained much attention in the last years especially in the field of systems biology. It has been recognized as a rich source of information extending the existing sources since it allows simultaneous collection of spatial and temporal protein information. In order to enable a high-throughput and high-content image analysis, sophisticated image processing routines become essential. We present a machine learning based approach for semantic image annotation i.e. identifying biologically meaningful objects. A semantic annotation becomes necessary, if image variables have to be associated to single biological objects, for example cells. We apply our method to pancreatic tissue sample images to detect and annotate cells of the Islets of Langerhans and whole pancreas. Based on the annotation, aligned multichannel fluorescence images are evaluated for cell type classification allowing accurate and rapid determination of the cell number and mass. This high-throughput analytical technique, requiring only few parameters, should be of great value in diabetes studies and for screening of new anti-diabetes treatments.