Deep Learning for Automated Microscopy Analysis in Biological Experiments
| advisor: | Ing. Adam Novozámský, Ph.D. |
| e-mail: | show e-mail |
| type: | phd thesis |
| branch of study: | MI_AMSM, MINF |
| key words: | artificial intelligence, image analysis, effect quantification, interpretability, LLM |
| link: | http://adamnovozamsky.com |
| description: | Comparative experiments represent a foundational paradigm of scientific discovery. Through systematic variation of input conditions and quantification of corresponding changes in experimental outputs, it is possible to identify relevant dependencies and formulate new hypotheses. Contemporary experimental platforms, encompassing high-throughput imaging and video-based systems, yield large and heterogeneous datasets. Manual analysis of these datasets is laborious, difficult to reproduce, and ill-suited for efficient exploration of the experimental parameter space.The objective of the dissertation is to develop innovative artificial intelligence methodologies for the automatic and robust quantification of effects in guided comparative experiments, with a primary focus on biological image and video data. The research will concentrate on two primary areas. First, it will address the extraction of expert-defined features from image and video data, including the segmentation of user-defined objects, tracking, and quantitative characterization. A key feature of this approach will be the minimal need for annotated training data and model fine-tuning. Second, the research will focus on the automatic discovery of new quantifiable features in biological datasets. The dissertation will further address methods for the visual amplification of detected effects, with an emphasis on interpretability for expert users in the biological sciences. Another objective will be the modeling of relationships between experimental inputs and outputs using probabilistic and neural-process-based approaches, together with the design of efficient sampling strategies for the exploration of the experimental parameter space. A secondary objective will be to leverage large language models to facilitate intuitive querying of experimental data and to enable interaction between the user and the experimental loop. The topic will be developed in collaboration with biological and biochemical research groups. |
| references: | 1. Kirillov, A. et al. Segment anything. In: Proceedings of the IEEE/CVF international conference on computer vision. 2023. p. 4015-4026. DOI 10.48550/arXiv.2304.02643 2. Carion, N. et al. SAM 3: Segment Anything with Concepts. arXiv. 2025. DOI 10.48550/arXiv.2511.16719 3. Woo, S. et al. ConvNeXt V2: Co-designing and Scaling ConvNets with Masked Autoencoders. arXiv. 2023. DOI 10.48550/arXiv.2301.00808 4. Siméoni, O. et al. DINOv3. arXiv. 2025. DOI 10.48550/arXiv.2508.10104 5. FRAZIER, Peter I. A Tutorial on Bayesian Optimization. arXiv [online]. 2018. DOI 10.48550/arXiv.1807.02811 6. Lipman, Y., et al. Flow Matching for Generative Modeling. arXiv preprint arXiv:2210.02747, 2022. 7. Rasmussen, C. E., Williams, C. K. I. Gaussian Processes for Machine Learning. MIT Press, 2006. 8. Garnelo, M., et al. Conditional Neural Processes. Proceedings of the 35th International Conference on Machine Learning, 2018. 9. Hernández-García, A., et al. Multi-fidelity Active Learning with GFlowNets. arXiv preprint arXiv:2306.11715, 2023. 10. Müller, S., et al. Transformers Can Do Bayesian Inference. arXiv preprint arXiv:2112.10510, 2021. 11. Boiko, D. A., MacKnight, R., Gomes, G. Autonomous Chemical Research with Large Language Models. Nature, 624(7992), 570–578, 2023. |
| note: | co-supervisor - prof. Ing. Filip Šroubek, Ph.D. DSc. The specific application focus is expected to be refined during the initial phase of the PhD study. |
| last update: | 20.04.2026 09:39:40 |
administrator for this page:
Ľubomíra Dvořáková | last update: 09/12/2011