Background:
Liver transplantation is a last-resort solution for patients with end-stage liver failure. However, due to the shortage of donor organs, extended-criteria donors (older, steatotic, or ischemic) are increasingly used. While these grafts are essential for reducing waiting lists, they carry higher post-transplantation complication risks, especially when affected by liver steatosis. Steatosis is characterized by fat accumulation in hepatocytes and is classified into microsteatosis (small droplets) and macrosteatosis (larger droplets), with the latter being particularly critical for graft viability.
Currently, steatosis evaluation primarily relies on biopsies and histopathological analyses—an invasive, subjective method that provides only localized liver assessments. To provide a faster and non-invasive alternative, near-infrared diffuse reflectance measurements have recently been performed on liver samples as part of Antoine Uzel's thesis. This optical method enables the collection of information on tissue properties without the need for a biopsy, offering a promising alternative.
Internship Objectives:
This internship aims to develop a non-invasive method to assess hepatic microsteatosis and macrosteatosis using Monte Carlo simulations and a deep learning algorithm. The intern will participate in all stages of the project, from theoretical modeling to experimental validation.
1. Monte Carlo Simulations:
Use Monte Carlo simulations to model diffuse reflectance in the liver under various microsteatosis and macrosteatosis conditions. These simulations will provide a database of infrared spectra correlated with liver steatosis levels.
Key Objectives:
- Model diffuse reflectance under different steatosis conditions.
- Study the impact of lipid droplet size and density variations.
2. Deep Learning Model Training:
Train a supervised deep learning model, using Graph Neural Networks (GNN), on the spectra generated by the simulations to predict steatosis levels. The algorithm should be able to distinguish between microsteatosis and macrosteatosis by analyzing the infrared spectra.
Key Objectives:
- Train a model based on simulated data.
- Optimize the model to improve the accuracy of steatosis predictions.
3. Experimental Validation:
Test the model on real spectra acquired from biological samples (mouse or phantom) and compare the results with histopathological analyses to validate the method's reliability.
Key Objectives:
- Compare algorithm results with histopathological reference data.
- Option to acquire in vivo data (mouse/human graft) or phantom data if desired by the candidate.
Candidate Profile:
The candidate should have strong skills in numerical modeling and data processing, as well as knowledge of machine learning algorithms. Previous experience with Monte Carlo simulations, spectral analysis, or biomedical optics would be a plus.
- Programming knowledge in Python (experience with PyTorch or TensorFlow is a plus).
- Knowledge of MATLAB would be an asset.
- Interest in biomedical optics and deep learning.
Skills Developed During the Internship:
- Mastery of Monte Carlo simulation techniques for light-tissue interaction modeling.
- Advanced skills in deep learning and biomedical data processing.
- Analysis and processing of experimental data from infrared spectrometry.
Additional Information:
- Duration: 4 to 6 months
- Location: CREATIS Laboratory, 21 Avenue Jean Capelle O, Villeurbanne
- Supervision: Cédric Ray-Garreau (CREATIS), Bruno Montcel (CREATIS), Antoine Uzel (CREATIS)
- Application: Send CV, transcript, and cover letter to cedric.ray[at]univ-lyon1.fr, bruno.montcel[at]univ-lyon1.fr, and antoine.uzel[at]creatis.insa-lyon.fr
[1] A. Uzel, M. Sdika, S. Chopinet, O. Lopez, et B. Montcel, « Near infrared diffuse reflectance spectroscopy for fat quantification in non-alcoholic fatty liver disease », in Diffuse Optical Spectroscopy and Imaging IX, SPIE, août 2023, p. 117‑120. doi: 10.1117/12.2669268.
[2] A. Uzel, M. Sdika, S. Chopinet, O. Lopez, et B. Montcel, « Near infrared diffuse reflectance spectroscopy for fat quantification in fatty liver disease », in Advanced Biomedical and Clinical Diagnostic and Surgical Guidance Systems XXII, SPIE, mars 2024, p. 54‑61. doi: 10.1117/12.3001895.
[3] D. A. H. Neil, M. Minervini, M. L. Smith, S. G. Hubscher, E. M. Brunt, et A. J. Demetris, « Banff consensus recommendations for steatosis assessment in donor livers », p. 12.
[4] R. Nachabé, Diagnosis with near infrared spectroscopy during minimally invasive procedures. 2012. Consulté le: 6 septembre 2023. [En ligne]. Disponible sur: https://repub.eur.nl/pub/32630/
[5] S. A. Prahl, « A Monte Carlo model of light propagation in tissue », présenté à Institutes for Advanced Optical Technologies, G. J. Mueller, D. H. Sliney, et R. F. Potter, Éd., Berlin, Germany, janv. 1989, p. 1030509. doi: 10.1117/12.2283590.
[6] Wu, Z., Pan, S., Chen, F., Long, G., Zhang, C., Yu, P.S., 2021. A Comprehensive Survey on Graph Neural Networks. IEEE Transactions on Neural Networks and Learning Systems 32, 4–24. https://doi.org/10/ggrj8p
[7] Zhou, J., Cui, G., Hu, S., Zhang, Z., Yang, C., Liu, Z., Wang, L., Li, C., Sun, M., 2020. Graph neural networks: A review of methods and applications. AI Open 1, 57–81. https://doi.org/10/gjt96x