News
Dr.-Ing. Dorina Strieth receives research funding from the TU-Nachwuchsring
Dr.-Ing. Dorina Strieth receives € 7.890 from the TU-Nachwuchsring for the research project “Automated evaluation of OCT images”
In the currently running DFG project (STR 1650/1-1), we are investigating the use of phototrophic biofilms that can improve plant growth in the agricultural sector. Phototrophic biofilms consist of bacteria that photosynthesize like plants. We worked with cyanobacteria, which have the ability to fix nitrogen from the air and release it into the environment in the form of ammonium. To investigate this, phototrophic biofilms were cultivated together with plants. It was shown that the biofilms have a positive influence on plant growth. In addition, different dry periods were investigated and it was shown that in the pot cultures, the soil covered with biofilm contained more pore water than those cultivated without biofilm. This means that the biofilm actively retains water in the soil. For this reason, the influence of different dry periods on biofilm formation and the ability to store water in the biofilm as well as the speed of drying out was investigated. For this purpose, optical coherence tomography (OCT) was used, with which the layer thickness of the biofilms can be imaged non-invasively. The layer thickness decreases when the biofilms dry out and increases accordingly during rehydration. The decrease or increase in biofilm thickness during dehydration or rehydration can then be used to draw conclusions about how quickly the biofilm absorbs water or how long it takes to dry out completely. In addition, OCT can also be used to investigate biofilm growth over the cultivation period. This method is well established at the department and has been used for several years. However, up to 1000 OCT images are taken per drying experiment, which then have to be analyzed manually. In concrete terms, this means that each individual image is currently opened with ImageJ, the biofilm is manually circled and the average layer thickness is then read out. In total, this takes around 5 minutes per image and a series of tests comprises around 2000 images, which means that the evaluation alone takes 166 hours with full attention.
Various attempts have already been made to automate this evaluation. Due to the poor image quality resulting from the resolution of the OCT and the layer thickness of the biofilm as well as its pigmentation and the resulting signal noise, an automated evaluation is difficult without AI/deep learning, even after consultation with experts in image evaluation.
As part of the project, deep learning software for the automated evaluation of OCT images is to be developed together with a start-up from Karlsruhe (HS Analytics). This should guarantee a fast and reproducible evaluation of OCT images in the future.