FM4L

This project aims to develop personalized, multiscale digital twins to enhance the understanding, diagnosis, and treatment of a subtype of leukemia: acute myeloid leukemia (AML). AML is driven by specific somatic mutations which influence disease evolution and treatment response. AML occurs as the rare ultimate step of an universal aging process, age-related clonal hematopoiesis

(ARCH), where hematopoietic cells acquire mutations over time and contribute to a variety of non-malignant and malignant diseases.

Our project focuses on developing an Artificial Intelligence (AI) platform, powered by multimodal large language foundation models (LLMs) able to process our microscopy images, to provide personalized prognosis for leukemia patients, and prediction for pre-leukemic patients through comprehensive digital twin models.

By leveraging advanced photonic imaging techniques based on Fourier Ptychographic and STED fluorescence super-resolved microscopies and integrating detailed genomic and bioclinical data, we aim to develop a more comprehensive understanding of individual pathologies. Additionally, our platform will exploit temporal data series for tracking cell population dynamics and monitoring patients over time. This enables us to analyze changes in cell behavior and population shifts, offering valuable insights into disease progression and treatment effectiveness. This interdisciplinary approach draws on fields such as bioinformatics, biophysics, advanced imaging, and AI, effectively bridging the gap between cellular-level insights and patient-specific outcomes. By combining these diverse data types, our platform supports multiscale modeling that leverages AI tools—including specialized LLMs —for processing cytologist descriptions and clinical annotations.

With access to this extensive dataset, our platform provides a deep understanding of intracellular behaviors of leukemic and pre-leukemic cells and the long-term dynamics of hematopoietic cell populations. This in turn supports precise, personalized prognosis, moving beyond simple positive or negative predictions toward a more nuanced view that informs individualized predictions and treatment strategies.

Ultimately, the integration of advanced imaging techniques with AI-driven modeling enhances our understanding of leukemia at a molecular level, representing a breakthrough in predictive analytics.

By offering highly personalized insights and care solutions, our platform has the potential to transform patient outcomes. The project is structured around 5 key work packages, each focusing on specific aspects of data integration, AI model development, and clinical application:

1) Developing enhanced innovative microscopy

2) Multiscale Data Specification and Production

3) Image Classification and Cellular modeling of leukemia-related cells

4) Temporal Series Analysis for Patient Evolution and Prognostic Translation

5) AI-Driven modeling and Prognostic Tool Development

This project is led by teams working together since many years and more recently within the framework OT4D, of a joint laboratory funded by the ANR. It brings together with the industry two hospital clinical teams, a biology research team, a physical optics research team, each of these partners being familiar with AI tools and contributing their own resources.