Apply Deep Learning in Microscopy Image Analysis with IKOSA AI
IKOSA AI is KML Vision’s novel software solution employing state-of-the-art deep learning techniques in microscopy image analysis.
Never before has microscopy image analysis been so fast and targeted to your needs! Create robust and reproducible AI-driven models best suited to the needs of your research institution without any prior coding knowledge. Minimize user bias in image recognition with the help of custom-trained algorithms.
Learn more about IKOSA AI
Deep learning and artificial intelligence transform the discipline of virtual microscopy by applying trained algorithms on biomedical image data. The trained algorithms running in the AI engine of IKOSA AI rely on deep convolutional neural networks (CNN) to learn from sample images in training datasets. Later, they apply the learnt concepts when predicting other image data.
IKOSA AI allows end users from life science research institutions to create their own replicable algorithmic methods for the automated examination of microscopy image data. This high-end AI-driven tool enables users to train and validate their own deep learning-based algorithms and run them on different datasets.
Algorithm development for quantitative digital microscopy made easy
Algorithm training made easy with IKOSA AI. Develop powerful algorithms for the study of microscopy images in a few simple steps.
Learn more about deep learning algorithm training
From a simple framework to a running algorithm. Here is how it works. Developing AI models for biomedical image analyses using IKOSA AI involves the following stages: data upload, dataset preparation, algorithm training, training report review, (optional) retraining of the algorithm and model deployment.
In the initial stages of model development you have to draft the automated method having your research design in mind. Upload your input dataset of digital images to the IKOSA Platform. Check our FAQ section to find out how many images you need for training the algorithm. Prepare your training dataset by selecting representative images acquired from an optical microscope source or another imaging device.
Next, select the images you want to use for the training in IKOSA AI. Annotate regions of interest in your image files based on user-defined ontologies and assign labels to the annotations. Review the summary of labels and annotations assigned to your image data. Then, train the model on your input dataset using the IKOSA AI Trainer. Optimizing your model is done automatically with the help of advanced deep learning techniques and neural networks at work.
Later in the validation stage you can test and benchmark your deep learning model on your bioimage datasets. You can do this in a few steps: review the training report, retrain the algorithm, if necessary, and eventually run your automated method on image datasets to perform complex quantitative analyses.
For more information on algorithm training, interpretability of results and validation, please refer to our FAQ.
If you are interested in specific examples of deep learning algorithm training, visit our case study page.
View Our Guided Step-by-Step Algorithm Training Workflow with IKOSA AI
Benefit from IKOSA AI: Deep learning in microscopy image analysis makes a difference
Push bioimage analysis further using trained AI models. Add value to existing workflows in your lab and accelerate performance using one intuitive tool. Here is how introducing trained algorithms to your image analysis process can benefit your research endeavors.
Research portfolio boost-up with trained algorithms
By using deep learning networks that can learn from data and adapt to new datasets, you can take your investigations to the next level.
IKOSA AI now enables you to train your own deep learning algorithm, which best matches the specifics of your experimental methods.
Build up your research portfolio by using trained algorithms that allow you to replicate analytical processes across different experimental settings and datasets of images.
Outstanding image recognition capabilities
Using trained algorithms helps prevent human error and subjective bias when detecting and labelling objects in microscopy images.
Automated and standardized manual image analysis processes offer greater sensitivity in object detection and unmatched consistency of the analytical outputs.
No coding skills required
You don’t need to be a programmer to use our IKOSA AI deep learning solution. The user-friendly and intuitive interface of IKOSA AI allows you to train your own algorithm without prior coding skills.
Trained algorithms in microscopy imaging allow you to quickly and accurately complete complex analytical tasks. Being able to run trained algorithmic models on new datasets gives you and your team more flexibility.
Once developed, the algorithm can be used by individual members of your team and applied to new samples and use cases.
Leverage the power of deep learning in microscopy
Accelerate research outcomes with IKOSA AI. Simply contact us to request a demo or a one month free guided trial version!
Improve performance and add value to existing workflows
We are constantly improving our software, making it more flexible and robust to meet the diverse needs of our customers.
Ongoing learning support
Gain knowledge on deep learning in microscopy imaging with our education materials. Stay up to date about the latest AI techniques in biomedical imaging with our educational content.
We offer you access to technical support materials and professional consultation along the way on topics like implementing the IKOSA AI tool or developing your first pre-trained algorithm.
Smooth integration into existing frameworks
Algorithms trained with IKOSA AI can be easily integrated into different data management systems and other software to add extra functionality.
IKOSA AI supports multiple input image formats (JPEG, PNG, BMP, TIFF, VMIC, GTIF, SVS, NDPI, SCN, STK, QPTIFF) as well as standard data formats (CSV, JSON) for quantitative output data.
Free trial and flexible conditions
To unlock the full potential of deep learning simply contact us to enable IKOSA AI in your account.
If you opt for a 4-week guided trial period, you can benefit from dedicated workshops, demo videos and tutorials on topics like image annotation and AI-training.
Apart from that, we offer you ongoing support with personalized consultation on potential technical queries that you might have.
API and plug-in options
The IKOSA Platform comes with a standard RESTful API enabling the integration into both commercial and open source imaging solutions like Fiji/ImageJ or CellProfiler etc.
On request we provide software development kits (SDKs) for all standard languages. If you are interested, we offer to support you in the development of plug-ins tailored to your requirements.
Check our FAQ-section to learn more about IKOSA AI
Still wondering whether IKOSA AI is the best deep learning solution for your research project? Or you still need some answers regarding the built-in capabilities of IKOSA AI and how they can help you enhance research performance? Our team of AI and computational microscopy experts answers your burning questions.
All important whys and hows about algorithms
How does deep learning for the study of microscopy image data work?
Deep learning in biomedical imaging relies on trained neural networks in order to recognize objects and patterns, for instance in microscopy image data. In general, deep learning is a subfield of machine learning. The most frequently used neural network types for image analysis are convolutional neural networks (CNN), a multi-layer neural network model capable of capturing hierarchical representations of the image input very well.
Traditional machine learning methods essentially make use of user-defined parameters such as object shape and texture, area or intensity in order to quantify microscopy images. Similarly, neural networks can be trained using human-labelled areas of raw images as input. Hence, a deep neural network takes it one step further and includes feature learning as part of the training process.
This allows the network to autonomously adapt to novel datasets of images and research designs in a self-learning manner. Neural networks can learn patterns from biomedical image data and later apply this to perform complex assays. This capability of deep learning algorithms significantly reduces the need for human intervention when conducting analytical tasks and increases productivity.
Which analytical tasks can be automated with the help of deep learning?
Our deep learning microscopy software allows you to automate complex segmentation assays on your data and obtain unbiased and reproducible results. IKOSA AI can currently assist in training all sorts of segmentation models: from cell segmentation and nuclei segmentation to tissue segmentation.
If the algorithm training is human assisted does this mean that human subjective bias cannot be completely ruled out?
Neural networks learn from labelled training image data. A neural network recognizes objects and patterns based on labels assigned to the training dataset images by a number of domain experts on behalf of your team.
In the course of its training phase the model adopts the common consensus among those experts when analyzing image data. This means that its prediction is much more objective and accurate than is the case for manually annotated input images.
Can we use ready-made Prisma algorithms available on the IKOSA Platform?
Using all the ready-made AI-based applications available on the IKOSA Platform is possible, if they fit your experimental design and dataset well. If the existing applications are not well fitted to your data you might need to modify and fine-tune them. We offer you active support in the course of adapting one of our available tools to your immediate needs.
What data has been used for training the existing Prisma algorithms?
Each algorithm has been trained on images and labels provided by domain experts for specific use cases. Technical specifications of the input data and imaging modalities are provided in the documentation of each algorithm.
Can we retrain existing algorithms with IKOSA AI to better fit our research design?
You can retrain an existing algorithm by running a previously generated model on a new training dataset.
The retraining of a deep learning model ensures a better fit to the data and minimizes prediction errors.
Our team of experts offers you effective support and advice with regards to retraining already existing models.
Can we export ready-made algorithms from the IKOSA Platform to other locations?
Not currently, but an export option to the most popular machine learning software frameworks is foreseen in the future.
Can we import our own algorithms?
Not yet for retraining the algorithm on the IKOSA platform. However, if you have a trained algorithm and want to use it on the platform, this is possible with a little help from our team.
Who has access permission to the trained algorithm on the IKOSA platform?
Only you and your team members in an IKOSA organization have access to your trained models and you can explicitly define who has access to output data.
Additionally, the output can be downloaded and shared with members of your team outside of the platform.
Algorithms cannot be shared with other organizations to protect the confidentiality of your work.
What are the regulations for updating algorithms?
Once trained, the algorithms do not automatically update themselves using new data. You completely remain in charge of when you want to update the algorithm with new data.
All important whys and hows about training
How much time does it take to train an image analysis algorithm with IKOSA AI?
The time it takes to train a deep learning network with IKOSA AI varies depending on the specifics of the analytical task and the quality of the image input data.
On average it takes about 20 to 30 minutes to get a new algorithm going. After a day or two of retraining your custom-made algorithm acquires greater robustness and is application-ready for productive microscopy image analysis.
What data do we have to use to train the algorithms?
Neural networks are trained on biomedical image datasets for specific use cases. Users need to provide a small number of images that are representative of a larger dataset and contain the structures they want to include in the analysis.
How many images are needed to train the algorithm?
This varies depending on your research design and the quality of your input data. However, you should provide at least 5-10 input images to start with.
What if our training dataset does not include enough images?
IKOSA AI allows you to train an algorithm even on small bioimage datasets. Functional algorithms can be developed using datasets as small as 10 images.
How do we decide what images to include in the algorithm training?
During the training phase the neural network learns on the basis of all the image inputs you provide. It decides autonomously whether image properties like size, texture or color are most relevant for the given analysis process.
Select images that best represent your sample and contain variations of the morphological structures you want to examine. The more diverse input image data you provide, the more robust and accurate your algorithm will be.
How can we figure out what needs to be retrained?
When the algorithm makes mistakes, this often is related to omitting visual items during training. Try to find objects that were not recognized properly and add them to the next training iteration by labelling more of them in your images.
How does outcome validation work with IKOSA AI?
Benchmark the outcomes of your algorithm-based image analyses with the results of more traditional methods (i.e. manual analyses, rule-based system or semi-automated analyses) using an identical sample. Thus, you can assure yourselves of the unmatched image detection capabilities of AI in the field of microscopy.
How can we interpret the output results?
The documentation accompanying each algorithm provides information about each label included in the model. Both quantitative metrics and categorical qualitative outputs are provided to enable verification and plausibility checks.
The included visualizations assist the interpretation of the outcomes. In order to give you actionable insights for potential improvements, IKOSA AI displays both correct (i.e. “positive” outputs) and incorrect detections (i.e. “negative” outputs) made by the algorithm.
Within each algorithm, different processing steps produce intermediate outputs. First, the AI algorithm predicts an image by projecting certain objects in the image file as located within a certain confidence interval. Some objects may get filtered out from the final result, because of some post-processing properties e.g. small size or too little confidence (less than 50%).
You can select particular regions of interest (ROI) in the image, run the algorithms on them and then view the results with the assistance of the interpretable visualizations enabled.
Contact us to get started with IKOSA AI
Contact us to get a one month free guided trial version, ask questions or request additional information on the uses of IKOSA AI. We guarantee you active support and quality advice as you test and implement the IKOSA AI deep learning solution.