Endoscopic ultrasound video analysis for pancreatic cancer examination quality assessment
Since the last two decades, medical community observes a worldwide dramatic increase in pancreatic cancer cases. Even if not of high prevalence, pancreatic cancer is foreseen to be second cause of cancer-related death by one decade from now. This is generally due to a late diagnosis, which is related to the inability of standard medical imaging (CT scan, Magnetic Resonance Imaging) to detect lesions smaller than 20mm. Endoscopic ultrasound (EUS) stands as an efficient alternative with which experienced EUS practitioners can detect lesions as small as 5mm. This early diagnosis enables efficient treatments leading to a one-fold increase in 5-year survival reaching up to 80%. Unfortunately, expert-level EUS practice is complex which impedes its accessibility at scale. Our group aims at providing non-expert practitioners with live assistance in the examination room in order to democratize EUS diagnosis and make it accessible to a larger portion of the population.
The proposed Engineer position is focused on developing Deep Learning (DL) methods to assess the quality of an ultrasound-based pancreas examination. Because of the high anatomical complexity of the gland and its associated veinous/ductal trees, EUS video analysis is by nature complex. In the context of pancreatic tumor detection, this complexity is exacerbated because of the general degraded state of the patient’s pancreas gland. Under this constraint, we aim at developing Deep Learning based methods to detect lesions as small as 5mm in live examination scenarios. In continuity of our previous works [1, 2], we will explore state-of-the-art DL computer vision algorithms for this task. Nevertheless, US images are significantly different from natural images both in terms of acquisition mechanism and in terms of local properties. In consequence, we will also explore innovative DL-based methods natively considering such peculiarities. All approaches will be benchmarked on our clinical datasets. We will also consider using self-supervised learning to fully exploit all available EUS data.
The position is located in Strasbourg (France), a lively, green, and cosmopolitan city in the heart of Europe and is also home to the European Parliament. The successful candidate will work with the research group CAMMA (Computational Analysis and Modelling of Medical Activities) and be hosted within the IHU institute for image-guided surgery (Institut Hospitalo-Universitaire de Strasbourg) at the University Hospital of Strasbourg. The candidate will thereby have direct interactions with clinicians in an exceptional international research environment offering unique clinical facilities. We also have a strong industrial collaboration in this topic . The CAMMA research group already possesses a large EUS videos database of pancreas examination, from which a significant, even if not complete, portion is annotated. We also have dedicated computing power necessary for the development of the algorithm in addition to a privileged access to examination rooms for live testing of the proposed algorithms in a relevant environment.
- Master’s degree in computer science or equivalent
- Strong knowledge of computer vision, machine learning and in particular deep learning
- Knowledge in ultrasound imaging and physics is a plus
- Proficiency in English (oral and written)
To apply, please send a long CV and a motivation letter describing your experience in computer vision, machine learning and medical image analysis to Jean-Paul Mazellier. The initial position duration is 12 months and salary will be commensurate to the profile and experience of the candidate.
 Meyer, A., Fleurentin, A., Montanelli, J., Mazellier, J.- P., Swanstrom, L., Gallix, B., Exarchakis, G., Sosa Valencia, L. & Padoy, N. (2022). Spatio-Temporal Model for EUS Video Detection of Pancreatic Anatomy Structures. In Simplifying Medical Ultrasound: 3rd International Workshop ASMUS 2022, held in conjunction with MICCAI 2022.
 Fleurentin, A., Mazellier, J. -P., Meyer, A., Montanelli, J., Swanstrom, L., Gallix, B., Sosa Valencia, L. & Padoy, N. (2022). Automatic pancreas anatomical part detection in endoscopic ultrasound videos. Computer Methods in Biomechanics and Biomedical Engineering: Imaging & Visualization.