Filamentous fungi is present in all types of ecosystems around the world. They mostly develop as a network of cells called hyphae. Organelles called vacuoles are responsible for nutrient storage as well as degradation of redundant organelles. They secrete primary metabolites for own metabolism while secondary metabolites contribute towards defense (antibiotics) or attachment (hydrophobins). Because of this, various fungal species are very important in fields like biotechnology, medicines, food production, and enzyme manufacturing, among others.

The amazing AI-based analysis solution provided by HS Analysis GmbH is based on strategies that incorporate objective analysis, which uses facts and evidence to draw conclusions from data without the use of subjective viewpoints or preconceptions, ensuring consistency and repeatability.

Preview of hyphae and vacuole detection in HSA KIT: A clear identification of viable structures is made possible by this advanced software which enables object detection in different complex spatial transformations

However, just identifying and analysing specific structures in a complex procedure might not be enough in cases like fungi fermentation. Despite having numerous industrial uses, the process from cultivation to product extraction is quite laborious because the overgrown hyphae not only obstructs the sensors and lead to unreliable measurement results, but also makes the broth more viscous.

Viscosity increases when the process is scaled-up from lab-scale to industrial processing; the mixing is heterogenous. Inconsistent aeration gives rise to stagnant zones

The optimum response to this issue would be to analyze the intricate process using AI software that mimics the physical process’s blueprint and provides the precise controlled conditions required for the operation. A digital twin would enable the seamless integration of several processes that affect the outcome. Furthermore, real-time data from sensors and other sources can be monitored to give a complete picture of the industrial process. Operators can spot abnormalities, pinpoint inefficiencies, and optimize the process parameters for increased performance, energy efficiency, and cost effectiveness by comparing the real-time data with the virtual representation of the digital twin.

Digital Twin: A virtual replica or simulation of a real system, procedure, or item. Predictive maintenance is made possible by examining historical and real-time data from sensors built into physical assets.


The HSA KIT software has an intuitive user interface and high-end professional annotation features to assure accuracy down to the last pixel. By standardizing processes, maintaining consistency, and promoting reproducibility, this can enhance the overall analysis process.

The software focuses on object detection and recognition which are fundamental functions of standard image processing. Using image analysis, object detection algorithms identify and categorize things of interest, providing useful data for further analysis and decision-making.

The FungiLyzer module focuses on detection of two most important structures for the study of fungal fermentation:

  1. Hyphae: Long thread-like structures (Red)
  2. Vacuoles: Spherical structures; either detached or within hyphae (Blue)

Zoom-out view

Hyphae are long, branched and filamentous structures
Vacuoles are membrane-bound, spherical structures

Zoom-in view

Absolute precision in the detection of both structures. The object and background are readily differentiated by the model because of the pixel-level classification training which delineates pixels within an image

True Negatives

Hyphae that have a white area inside are not of interest as they may be growing under sub-optimal conditions, and hence are less metabolically active.
Vacuoles consist of cell sap which appears to be clear or translucent, hence some white space is expected in the middle. An „all dark“ structure can indicate, for example Autophagy.

There also might be visualisation problems associated with the appearance of the structures. Hence, for a better training, such structures would not be detected by the model.

This is a zoom-in representation of the FungiLyzer model that rejects background structures and only recognizes the necessary structures. The software is designes to denoise and enhance image resolution to produce immensely accurate object detection.

Below, an area of a much larger image that demonstrates the true outstanding object detection quality that HSA KIT is capable of achieving is shown. The images below may be zoomed in and out, and there is a slider that can be moved from left to right to improve and enhance the viewing experience.

An example of detecting hyphae and vacuoles.

Zoom Out view of the whole image.

True positive hyphae

True positive vacuoles

After the model has been trained and the project has been completed, the HS Analysis software provides the information required to understand the results in a sheet format, including the name of the files, the number of objects (structures) such as hyphae and vacuoles, the dimension of the analyzed areas in square meters within specific Base ROI (Region of Interest) area.

The results sheet contains, but not limited to the basic information about the structures. In addition to this information, other details like average area, perimeter, minimum and maximum diameter etc.

While a basic model provides a general or broad overview of the subject matter, giving a high-level understanding without delving into the specific details or variations within the structures; the more customized module takes into account the diverse subtypes within each structure, ensuring that the results obtained are more refined and accurate.

In the image below, hyphae are further characterised into three classes:

  • Without vacuole: green
  • With vacuole: orange
  • Complex: purple
Class Hyphae further characterised into three sub-classes. This provides an overview of the ongoing fermentation, and helps to assume or identify what stage has the process reached

Detecting Hyphae with three sub-classes.

Biopesticides: Real world Application

The naturally occurring fungus are the source of biopesticides, which have been shown to be successful in lessening the damage that plant diseases, pests or weed do to crops. Since they don’t leave toxic residues in the soil or water sources, fungus biopesticides are also a more environmentally friendly option than chemical pesticides. Moreover, the process of fermentation involves the use of renewable raw material i.e. glucose.

FungiLyzer may detect any anomalies or differences in the fermentation process by evaluating images of the process, enabling for early intervention and prevention of potential inconsistencies in the process. This not only improves workflow efficiency but also minimizes the danger of economic losses for companies. FungiLyzer also delivers real-time data and insights, allowing for informed decisions and the optimization of culture procedures for maximum efficiency.

Workflow with HSA KIT

Analysing samples and digitalisation of the slides have never been easier before. HSA KIT provides an unparallel experience to its customers, who want to keep up with todays‘ „better“ alternatives and achieve higher efficiency in their workflow. The HSA team does all that is possible to please its clients, from the installation and integration of the software to the limitless support and upgradation.

AI based analyses using HSA KIT would provide:

  • Standardised process with subjective/objective analysis
  • Extraction of relevant features from raw data and create meaningful representations for training AI models
  • Module selection and configuration without excessive coding
  • Easy to learn software: Annotate , Train and Automate
  • Quick and efficient analysis of multiple medical images which cuts down time for diagnosis or treament
  • Automated report generation to boost productivity and help physicians or radiologists in evaluation process

We understand the importance of having accurate and efficient slide analysis software. That’s why we offer a range of features to help you achieve your goals.

  • Our software is compatible with both Linux and Windows operating systems, and it runs smoothly in Docker. With offline functionality, you can work on your slides without an internet connection, but if you need to extend to online use, that’s possible too.
  • We also offer professional integration into your network infrastructure, so you can seamlessly incorporate our software into your existing systems.
  • Our software is designed to work with other programs as well, ensuring that you have all the tools you need at your disposal.
  • Even on weaker computers, our software delivers full performance, so you can analyze slides quickly and accurately.
  • And with the ability to generate reports in CSV and Excel formats, you’ll have all the data you need at your fingertips.
  • Finally, user profile management ensures that each user has their own history and preferences stored for easy access.

AI based analyses can be way more efficient, however it can only act as a supportive tool to human expertise and never be a complete alternative. AI algorithms have the tendency to produce false positives and false negatives, and may have trouble adapting to a data that deviates from training dataset.

With our advanced technology and expert team of professionals, we are confident in our ability to deliver top-notch slide analysis services that meet the unique needs of each individual client. Whether you’re looking for a one-time analysis or ongoing support, we have the tools and expertise necessary to help you achieve your goals.

Better than alternative softwares

Strong numerical computing programmes can be used for data analysis, modelling, and simulation. Despite the fact that they have a wide range of industrial uses, there are a number of reasons why they might not be suitable for predicting scale-up in industrial processes.

  • Restricted scalability: Most software are not built to handle large-scale industrial operations; rather, they are best suited for small- to medium-scale tasks. For many industrial applications, scaling up the problem size can soon become computationally expensive and time-consuming.
  • Lack of industrial-specific tools: Despite all-purpose tools for data analysis, modelling, and simulation; it might not, however, have the precise models and techniques required for anticipating scale up in industrial processes, such as those associated with equipment design, safety analysis, and process optimisation.
  • Restricted software integration: Industrial processes frequently require a variety of software tools and platforms, and not all work perfectly with programmes and systems already employed in the sector. The modelling and simulation process may become inefficient as a result, making it challenging to anticipate scale up precisely.
  • Inadequate support for parallel computing: A single machine might not be able to handle the massive quantities of computational power needed for many industrial processes. Some parallel computing capability might not be enough to meet the needs of simulations used on an industrial scale.