AI Imaging Lab: Development & Validation of Segmentation and Detection Models
Service Description
Overview
This service supports SMEs and researchers in developing, training, and validating AI models for medical imaging applications. Hosted at SMAILE, Karolinska Institutet, it covers segmentation, detection, and classification tasks using CT, MRI, nuclear images, multimodal images, ultrasound, histology, and microscopic images. The pipeline includes data curation, evaluation of labeling strategy, model selection and training, evaluation metric selection, and model performance benchmarking.
We provide expertise and support in:
- AI model training and optimization for medical imaging
- Validation using clinical datasets and standard metrics (using publicly available datasets, datasets available through data agreements, and internal datasets at KI, depending on the case)
- Clinical Relevance and Comparison with the state of the art in research and clinical practice
- Imaging biomarkers studies for diagnosis, prognosis, and prediction applications
How can the service help you?
The service ensures your imaging AI model performs reliably and is aligned with clinical expectations. Whether you’re entering the pre-clinical testing phase or seeking validation to secure investment or regulatory approval, this service equips you with a rigorous evaluation and feedback report.
How the service will be delivered?
Available both virtually and physically. Imaging data can be reviewed remotely through a secure data transfer process. On-site collaboration is also possible for sensitive datasets or model development, evaluation, and validation. A typical project takes 4–6 weeks.
Additional information
Provider description
SMAILE is the digital health core facility at Karolinska Institute. It offers interdisciplinary support in AI for medical imaging, data analytics, and system validation, partnering with leading institutions under the Swedish TEF-Health node.
Technical description
The imaging model pipeline is extensively validated by using a curated benchmark set and standard evaluation frameworks such as well-established quantification metrics for object detection, image segmentation, and classification tasks.. Annotation quality is reviewed, and model performance is benchmarked against open or reference models. Standard medical image processing tools such as PyDicom, ANTs, and ITK, as well as community-driven open-sourced frameworks such as MONAI, are the core components of our designed pipelines.
Service customization
The service can focus on either model development, evaluation, or both. Datasets can be anonymized and securely shared, or analysis can be conducted in a local sandboxed environment.
Provider & Contact
Pricing is available to registered users. SMEs receive significant state-aid reductions (GBER) — or, depending on the call, free services during the funded project. Sign in or register to see the price for your organisation.
Sign in or register to see pricing