A.I.B.M. - Interpretable AI-Powered System for Identification of Bone Metastases in Computed Tomography and Magnetic Resonance Imaging

The proposed research aims to utilize the capabilities of Deep Learning (DL) to improve the accuracy of medical diagnostic imaging, specifically in the detection of metastases within computed tomography (CT) and magnetic resonance imaging (MRI). This study will employ 3D fully convolutional neural networks to analyze complex data patterns from medical imaging, focusing on identifying subtle anatomical structures associated with metastases in bone tissues. The project will develop a Deep Neural Network (DNN) with a modular architecture, incorporating convolutional layers and a diffusion denoising probability method to enhance the detection and characterization of dispersed metastatic lesions. These lesions present considerable detection challenges due to their size and variability. The network will be trained, validated, and tested on an annotated radiology dataset from Latvia, aiming to optimize the model’s performance for local populations and contribute to the broader application of DL in medical diagnostics. 

The project is divided into three phases:

1. The initial task is to improve medical understanding of bone metastasis detection while exploring the most effective deep convolutional neural network (DCNN) structures and algorithms. 
2. The second task focuses on dataset segmentation while conducting research and experiments using established DCNN and visual transformer deep neural network (DNN) architectures and algorithms. The initial model, capable of segmenting pelvic bones and spinal vertebrae, will be developed through deep neural network training. 
3. Validation of the deep neural network model in daily clinical practice, compared to the assessment of an experienced radiologist.

Outcomes

• Original scientific articles published, submitted or accepted for publication in the Q1 or Q2 quartile publications included in Web of Science Core Collection or SCOPUS databases - 1;
• Original scientific articles, published, submitted or accepted for publication in the scientific publications or conference symposia, included in Web of Science Core Collection or SCOPUS databases, for social, human and art sciences included in ERIH PLUS database - 2;
• Scientific databases and datasets developed within the project and prepared according to FAIR principles - 1;
• Other new product or technology, software copyrights (including methods, prototypes, treatment and diagnostic methods not to be commercialised, etc.) - 1;
• Other project results according to the specific nature of the project complementary to those listed above (including pre-prints);
• Presentation or poster at ECR 2025, RSNA 2025;
• Presentation or poster at ECR 2026, RSBA 2026, BCR 2026.

Project Research Team

• Maija Radziņa, Project Manager
• Viktorija Cīrule 
• Matīss Šņukuts 
• Laura Saule 
• Madara Ratniece 
• Roberts Šamanskis

Project Collaboration Partner

Institute of Electronics and Computer Science

Project-related activities

• Participation in ERASMUS+ Staff Week, Riga, Latvia, with the lecture From Data to Decisions: Implementing Artificial Intelligence in Clinical Practice. Principal Investigator Maija Radziņa presented the project’s approach to implementing artificial intelligence in clinical practice to an international academic audience. (02.06.2026.)
  Program of the event
• Oral presentation at the European Congress of Radiology 2026 (ECR 2026), Vienna, Austria, and publication of an electronic scientific poster in the EPOS™ database. The presentation Can a Limited Patient Sample Still Be Suitable in AI Training for Segmentation of Bone Metastasis on Computed Tomography (CT) Imaging? showcased the project's initial results in bone metastasis segmentation using artificial intelligence methods. (04.–08.03.2026.)
  Scientific poster  
• Participation in the international conference Innovation and Artificial Intelligence: The Path from Research to Practice during RSU Research Week 2025. Under the conference theme Bridging the Gap: From AI Research to Real-World Impact, Edgars Edelmers delivered the presentation Interpretable AI in Oncologic Imaging: The A.I.B.M. System for the Identification of Bone Metastases in Computed Tomography and Magnetic Resonance Imaging; Roberts Šamanskis presented AI Algorithms and Capabilities for Brain Perfusion; and Maija Radziņa delivered the presentation Modern Radiology in Tandem with AI: A New Path to Precision Medicine. (28.03.2025.)
  Conference program
• Participation in the 11th Baltic Congress of Neurology (BALCONE 2025). Professor Maija Radziņa delivered the lecture Neuroradiology and Artificial Intelligence: Opportunities in Latvia Today and Tomorrow, presenting the experience gained within the project in applying artificial intelligence to medical image analysis and disseminating project results to an international professional audience. (08.11.2025.)
  Conference program
• RSU Scientist's breakfast 04.06.2026.
  RSU tenured professor Maija Radziņa introduced the A.I.B.M. project, which is developing interpretable artificial intelligence solutions for identifying metastases in medical images of CT and MR. The system uses deep learning algorithms and annotated radiology data to improve diagnostic accuracy and reliability
  RSU Scientists’ Breakfast highlights current projects in AI and digital medicine

Publications

• Sudars, K., Edelmers, E., Ņikuļins, A., Cīrule, V., Šņukuts, M., Ratniece, M., Šamanskis, R., Sprūdža, K. L., Radziņa, M. Comparison of Foundation Models MedSAM and DINOv3 with nnU-Net Framework for Bone Metastasis Segmentation in Computed Tomography Scans. Journal AI, 2026, 7(6), 181. (04.06.2026.)  

AI is an international peer-reviewed journal publishing relevant, impactful and innovative research in artificial intelligence, as well as reviews and commentaries that help readers understand complex scientific innovations and their practical applications.

Project-Related Theses and Dissertations

• Madara Ratniece and Matīss Šņukuts. Residency research thesis Segmentation of Skeletal Metastases in CT and the Role in AI Model Development, defended on 14.08.2025.  

Radiology residents Madara Ratniece and Matīss Šņukuts presented research results on skeletal metastasis segmentation in computed tomography images and its role in the development of artificial intelligence models.  

• Edgars Edelmers. Doctoral thesis Automated Morphological Structure Detection, Segmentation, and 3D Model Reconstruction from Medical Images Using Artificial Intelligence Deep Neural Network Technologies, defended on 29.12.2025.  

The doctoral thesis focuses on the application of artificial intelligence methods in medical image analysis, including image segmentation, morphological structure detection and 3D model reconstruction, and is closely related to the scientific objectives of the project. 

Media

• This Isn't Fantasy: Real Research That Could Soon Make a Major Difference in Treating Serious Diseases, (article in Latvian) Delfi.lv (9 Jul 2026)
• What’s New in Modern Radiology? Maija Radziņa on the Use of Artificial Intelligence in Radiology, TV24 programme Dr. Apinis (in Latvian language) (08.09.2025)

Professor Maija Radziņa discussed the role of artificial intelligence in radiology, highlighting the opportunities and challenges of AI-assisted medical imaging and its potential applications in clinical practice.