AI-Powered Imaging
Platform for Precision
Medecine

Images are collected and stored on a secure platform based on secure File Transfer Protocol (sFTP), these images are then optimally organized for efficient and accurate processing.

Linked to secure storage archives, Ikonics streamlines image collection and allows us to characterize and structure medical images for every patient according to:

1. Imaging modality(CT, PET, MRI, etc.)

2. Acquisition and reconstruction parameters (kernel version, sequence, contrast phase, etc.)

3. Timepoints (previous treatment, baseline, follow-up, adverse event, end of treatment)

The quality of medical images is assessed based on eligibility criteria conducive to effective analysis.

Building a comprehensive, structured and readable imaging database is required to correctly process data. Linked to a secure storage archive, Ikonics® consolidates the image collection process and centralizes image quality assessment for enhanced efficiency. Perfect characterization and assessment are vital to defining what analysis and insights can be derived from each element of data. During this step we perform:

1. Anonymization check: We verify and ensure that no personal identifiable information (PII) is included in the transferred data.
2. Technical & usefulness check: We identify image resolution, image modality, sequence, reconstruction, and the comparability of body regions.
3. Medical review: Our medical team manually validates the quality of the previous steps i.e., organ completeness, absence of artefacts, and phase accuracy, as well as guiding organ and timepoint selection.
   
   
   

Imaging segmentation is the delineation of the region of interest, portioning the pixels in an image into district regions to assist in image processing tasks. This process can be manual, automated or semi-automated.


The segmentation process is one of the most critical steps of radiomic analysis. To ensure the process is completed to the highest standard, we have an internal team of annotators and medical professionals, as well as a network of radiologists, experts and privileged partners. Furthermore, all segmentations at radiomics.bio are validated by our medical team through an intensive quality assessment process (human-in-the-loop).

During this step, Ikonics® allows us to combine automatic AI image segmentation models and manual reviews into an efficient workflow, accurately identifying and tracking regions of interest (ROI) over time.

These ROIs can include whole organs, oedemas, necrosis, vessels, and abnormalities such as reticulation, ground glass opacity in the lung and diffused disease. Ikonics® also integrates both pre- and post-image processing tasks, giving complete control of the image feature quantification process for an optimal and efficient delivery. 

The lesions present in scans are given consistent labels to ensure accuracy. Lesions are linked with different timepoints in a process called “lesion tracking,” in which semantic links are established between regions of interest.

Lesion tracking is the linking of lesion present in different timepoints. With Ikonics® we can accurately identify and track regions of interest (ROI) over time. Consistent labels are given to the lesions present in the scans to ensure accuracy.

The inclusion of relevant radiological findings is also performed at this stage which is incorporated into patient specific reports for a comprehensive assessment of the medical image. Finally, post-processing steps are implemented to ensure perfect quality masks for the quantification of imaging features.

Radiomic features are extracted from the medical scans, each linked to specific characteristics describing the ROI over time in relation to response or lack thereof to treatment, Radiomic features belong to different categories, each of them quantitatively describes single characteristics of organs or lesions under examination, representing the size, shape, texture, and intensity of the ROI. Radiomic signatures are a combination of radiomic features used inside statistical models to predict or classify a specific condition or clinical endpoint of interest.

In this step radiomic features are extracted and quantified in order to describe the ROI at baseline and at multiple timepoints to understand how the lesion evolves following treatment. For research and healthcare applications, radiomics.bio uses machine learning algorithms to ensure explainability of results and provide confidence on the underlying biological mechanisms at play.

The final step in the radiomic workflow converts statistics to insights, ensuring that the quantified statistics are in line with biology.

We provide a wide range of deliverables with a variety of levels of complexity:

1. Case studies of patients of interest with high quality visual renderings
2. Fundamental statistical analysis e.g. basic statistics, linear regressions, etc.
3. Advanced statistics e.g. statistical distribution test, multivariate analysis,
4. Bayesian approaches, etc.
5. AI modelling such as machine learning, linear and non-linear effect models, and deep learning algorithms
6. Medical software development following our ISO 13585 certification

We review the relevance of the results obtained and how it correlates to biology including how the results relate to drug mechanisms, patient indications, the localisation of the tumour and tumour mutations etc., providing actionable information for insight-based decision-making