Skip to main content



The mission is to increase recognition for in-vitro and in-chemico tests by digitalizing the biotechnology field. Cell-Lab should provide software for testing the toxicitity of pharmaceuticals, medical devices and cosmetics by analysing microscopy images. The goal is a prototype of a cell analysing software that follows the ISO 10993-5: Biological evaluation of medical devices - Part 5: Tests for in vitro cytotoxicity standard. First, images are analysed with traditional computer vision algorithms and prepared for training a neural network. Either U-Net, Mask R-CNN or YOLO will be considered for the automated feature extraction and segmentation.

Computer Vision Results

Edge Detection and Morphological Operators (MATLAB)

Basic edge detection uses the threshold of the derivative. More advanced edge detectors are the Sobel operator and Canny edge. The two basic binary morphology operators are dilation and erosion, and are the base for opening and closing. Dilation makes the foreground objects bigger, erosion whittles the objects down. Opening is a combination of the two basic operators, removing small objects from the background. In contrast, closing removes small holes in the object.

Contours (OpenCV/C++)

Active contours is a way to segment an image. First, a circle is drawn outside an object. The contour will then move until it encounters the object. Different approaches are made to realize the algorithm: level sets, and particle-based methods like snakes. OpenCV uses a version proposed in 1985.

K-Means with 4 classes (OpenIMAJ/Java)

Clustering is a technique used to segment an image. One of the most popular algorithms is K-Means, where similar colours are grouped together. The number of different classes needs to be defined.

Circular Hough Transform (MATLAB)

The used algorithm, based on the Circular Hough Transform is robust against noise, occlusion and brightness gradients. First, pixels with high gradients are assigned the status \textit{candidate pixels}. It is worth noting that gradients are also used by edge detectors. Around each candidate pixel, a circle with a defined radius is drawn. Each pixel of the newly drawn circles gets a "vote" which is saved in the so-called accumulator array. The array counts the votes together for each pixel, resulting in peaks in the middle of round objects. If multiple radii are defined as a range of numbers, the 3D Hough space can be used, where each circle is then visualised as a cone.

Marker-controlled watershed algorithm (OpenCV/C++)

The segmentation technique uses a set of markers, where the local minima are flooded. Grayscale images can be viewed as topographical surfaces, where a local minimum is the bottom of a lake. By pouring water with different colours on to the positions of the local minima, each lake starts to rise. When two different colours of water meet, a border is found which delimits the numerous objects.

Felzenszwalb Huttenlocher Segmenter (OpenIMAJ/Java)

The Felzenszwalb Huttenlocher algorithm is a graph-based image segmenter. It is based on pairwise region comparison, where the minimum weight edge between two regions is considered.

Events and Awards

Vienna Techstars Startup Weekend Women 2019

  • Attended mentoring and workshops
  • Pitching Cell-Labrity 4th supporter team for founders, freelancers and makers

  • 4 months programme
  • Attended workshops: website check, graphics check, social media check, funding check

MICCAI 2020, LABELS workshop

Team: Biotechnology meets Computer Science

So who’s behind Cell-Lab? Four woman from OFI, the Austrian research institute for chemistry and technology. The researchers are Elisabeth Mertl, Angelika Wlodarczyk, Martina Helmlinger and Elisa Mayrhofer provide cell images and help with life science related questions. The data scientist is Christina Bornberg, who is doing her Bachelor project about image analysis in the field of microscopic image analysis.