Mouse2Dat

The field of nuclear medicine holds great promise in combating solid tumours with poor prognoses and limited treatment options. In the preclinical development of diagnostic and therapeutic radiopharmaceuticals, in vivo biodistribution, along with receptor binding, forms the cornerstone for medicinal chemistry optimization. Presently, the evaluation of in vivo imaging studies is a manual and imprecise process, consuming substantial time and acting as a limiting factor. To enhance efficiency, there is a pressing need to automate the analysis of imaging data. This step will expedite the development of new medications, improve data reliability, reduce the need for animal experiments in individual projects, and ultimately bolster the competitiveness of 3BP as a successful player in the field of nuclear medicine.

In nuclear medicine, targeted radioactively labelled probes, known as tracers, are employed for diagnostic imaging, which can also be utilized for therapeutic nuclear medicine. These tracers contain a binding element that specifically and with high affinity binds to receptors primarily or exclusively expressed in tumours. For therapeutic use, a radionuclide is employed, emitting high-energy particles (electrons or alpha particles) that destroy cancerous tissue.

3BP has been developing innovative radiopharmaceuticals for over a decade, focusing on cancer types with a significant medical demand for more effective and less side-effect-prone treatment options. During lead structure optimization, 3BP synthesizes and tests between 50 and more than 1000 different substances in iterative cycles, typically several hundred.

Currently, the evaluation of data from small animal imaging studies is manual, exceptionally time-consuming, imprecise, and lacks reproducibility. There is a strong need for a highly automated bioinformatics evaluation solution.

To significantly enhance the efficiency, reproducibility, and objectivity of evaluating data from small animal imaging studies, an automatic evaluation tool for this data is being developed in collaboration with the project partner Fraunhofer Heinrich Hertz Institute (HHI).

AI-Enhanced Segmentation: HHI employs AI methods, including state-of-the-art machine learning and image processing techniques, to implement automatic segmentation of CT and SPECT data. This segmentation is crucial for isolating regions of interest for analysis. The distribution and density of the tracer agent within these segmented areas will be evaluated using statistical methods.

Automated Grading with Active Learning: To further enhance the efficiency and accuracy of the segmentation process, we will employ active learning methods and uncertainty estimation. This will allow to automatically grade segmentation results, providing insights into the certainty of these results, highlighting instances where manual correction may be necessary.

Data Input and Output Framework: We are constructing a versatile software framework capable of receiving datasets alongside information about injected radioactivity doses. This framework will conduct an in-depth analysis of radioactivity biodistribution. Crucially, it will provide users with results of individual steps within the evaluation process for manual review, quality control, and potential correction. This feature ensures that experts at 3BP maintain oversight over the evaluation process.

• 3B Pharmaceuticals
• Fraunhofer HHI