Co-funded by the European Union (ERDF)


Intelligent Radiological Assistant (IRA)

The aim of the joint project of PLANET AI and the University of Rostock is to develop an Intelligent Radiological Assistant (IRA) that uses medical image data to suggest diagnoses. The type of disease is left open and only specified later based on training data. Thus, the product can easily be expanded to other diseases.

In a subproject, PLANET AI will develop a front-end workflow for IRA. An essential component is the training and evaluation environment including, among other things, data preparation, classification of the data into benchmarks on which to measure the development progress, and the implementation of suitable metrics to measure and analyze the performance.

PLANET AI’s image analysis capabilities provide the foundation for this project. State-of-the-art image classification and image segmentation technologies are evolving to provide results with the highest precision possible. Further, we will compare, transfer and develop state of the art neural models for medical 2D and 3D image classification.

The project duration is from 2020 to 2023.

Automatic Information Extraction from Documents (AID)

Requests, informal letters, invoices, certificates and receipts – the sheer number of documents in a company binds resources. Letters, parcels and applications must be read and forwarded. Important sender and concern information is manually transferred to databases before an agent can deal with the actual content.

To avoid such overhead, PLANET AI is developing an assistance system for document processing based on deep artificial neural networks in the “Automatic Information Extraction from Documents (AID)” project. A complex task, such as information extraction, must combine textual and visual features to achieve optimal results.

PLANET AI’s award-winning handwriting recognition and layout analysis technologies provide the foundation for these features. State-of-the-art Natural Language Processing technologies are evolving to combine features and extract information with high precision. The customer himself will be able to train and adapt systems to his needs without having to specify complicated rules or templates. The elaborate data production is to be avoided as much as possible and the assistant can learn new classes and features from a few sample data.

The project duration was from 2019 to 2021.