PhD Position F/M Quantitative analysis of X-ray angiography images in acute ischemic stroke

Ischemic stroke is a major cause of disability and death worldwide. The occlusion of blood vessels results in a potentially large area of the brain to be oxygen deprived. Recent clinical trials have demonstrated Endovascular Thrombectomy (EVT) to be highly effective, which has led to its widespread adoption in clinical routine. In the vast majority of cases, EVT is successful in the sense that recanalization is observed : the blood flow is restored in the previously blocked blood vessels, which can be assessed using Digital Subtracted Angiography (DSA). However, even when treatment has been performed within the recommended 3 hour-window after stroke onset, it is estimated that about 50% of successful recanalizations do not result in a restored perfusion in the downstream brain tissues. This phenomenon is known as no-reflow. But the reperfusion status cannot be readily assessed in DSA images, requiring a clinical evaluation of the outcome. Trustful quantitative tools are missing to assess the level of reperfusion after EVT, better predict potential difficulties for this reperfusion to be restored for a patient, and help the medical community better understand the underlying mechanisms of successful reperfusion. The objective of this project is to leverage recent deep learning models to develop image analysis tools and design appropriate image-based metrics to enable an immediate assessement of reperfusion based on DSA images acquired before, during and after EVT.

The Tangram Inria team has had a long standing collaboration centered on interventional imaging and simulation with medical practitioners from the Interventional Neuroradiology department of Nancy University Hospital, one of the world leaders in interventional neuroradiology. IADI Inserm team also collaborates with this medical team on developing innovative MRI imaging and multimodal image processing. This project is part of new research collaboration involving all three teams, with the aim to better understand the no-reflow phenomenon and make progress towards a more personalized, more efficient treatment for ischemic stroke. This thesis will be co-supervised by Erwan Kerrien, from Tangram team, and Julien Oster, from IADI. The PhD student will have an easy access to the medical expertise of Prof. Benjamin Gory, and colleagues.

DSA images present as 2D sequences of projection images showing a contrast medium flow throughout the brain vasculature to highlight the blood path. It follows three phases : arterial supply, parenchymal where the brain tissue is perfused, and the venous drainage.

The first objective will aim at automatically separating these phases. Previous works have addressed this question using Long Short Term Memory or Gated Recurrent Unit neural networks used to segment scenes in videos in a supervised way [Su21,Mittmann22], or using unsupervised Independent Component Analysis [Haouchine21,Baur24]. These fundamentally lead to uncertain solutions since they classify whole frames, whereas, due to projection and different separating times in different parts of the brain, images where phases are mixed should be considered. Unsupervised generation of three separate phase sequences will be targeted.

The second objective will aim at providing quantitative information about the perfusion status in the patient’s brain. Only a few recent works [Kosior19] have experimented with more or less simple application of CT perfusion principles to generate 2D perfusion maps (Mean Transit Time, Time To Peak, Cerebral Blood Flow, Cerebral Blood Volume), but again these maps suffer from projection distortion. This step will develop methods to generate such maps from the parenchymal phase of a DSA sequence, using 3D CT or MRI perfusion maps as annotations. We also target the development of new DSA-specific grading, e.g. inspired by the Myocardial Blush Grade used in interventional cardiology to grade a patient’s perfusion state from cardioangiography sequences [van’t Hof98], as a complement to the current grading system  (TICI and ASPECTS scores).

The third objective will target the assessment of the collateral circulation of a patient. The extension of such circulation is considered to be an important predictor of the clinical outcome, but there is currently no efficient and robust way to evaluate it [Consoli23]. Here again a multimodal approach, combining MRI and DSA imaging will be pursued.

• [Su21] Su, R. et al. (2021). "autoTICI: Automatic Brain Tissue Reperfusion Scoring on 2D DSA Images of Acute Ischemic Stroke Patients," in IEEE Transactions on Medical Imaging, vol. 40, no. 9, pp. 2380-2391. https://doi.org/10.1109/TMI.2021.3077113.
• [Haouchine21] Haouchine, N. et al. (2021). "Estimation of High Framerate Digital Subtraction Angiography Sequences at Low Radiation Dose". In: de Bruijne, M., et al. Medical Image Computing and Computer Assisted Intervention – MICCAI 2021. MICCAI 2021. Lecture Notes in Computer Science(), vol 12906. Springer, Cham. https://doi.org/10.1007/978-3-030-87231-1_17
• [Mittmann22] Mittmann BJ, et al. (2022). "Deep learning-based classification of DSA image sequences of patients with acute ischemic stroke". Int J Comput Assist Radiol Surg. 17(9):1633-1641. https://doi.org/10.1007/s11548-022-02654-8
• [Baur24] Baur, K, et al. (2024). "Spatiotemporal Disentanglement of Arteriovenous Malformations in Digital Subtraction Angiography." arXiv preprint arXiv:2402.09636.
• [Kosior19] Kosior JC, et al. (2019). "Exploring Reperfusion Following Endovascular Thrombectomy". Stroke. 50(9):2389-2395. https://doi.org/10.1161/STROKEAHA.119.025537.
• [van’t Hof98] van't Hof, AW et al. (1998). "Angiographic assessment of myocardial reperfusion in patients treated with primary angioplasty for acute myocardial infarction: myocardial blush grade". Zwolle Myocardial Infarction Study Group. Circulation. 97:2302-6.
• [Consoli23] Consoli, A et al. (2023). "Unfavorable clinical outcomes in patients with good collateral scores following endovascular treatment for acute ischemic stroke of the anterior circulation: The UNCLOSE study". Interventional Neuroradiology. https://doi.org/10.1177/15910199231212519.

The candidate will pursue research activities in deep learning on image sequences and 3D images. Weakly supervised and unsupervised methods will be leveraged. Access to computing resources will be granted. Data will be provided by interventional neuroradiologists from Nancy University Hospital, who perform EVT on 100-150 patients with acute ischemic stroke each year.

The candidate is expected to participate in meetings, report on his/her research, interact with other members of both Tangram and IADI teams, and in particular work in close collaboration with the involved interventional neuroradiologists to validate new ideas and ensure their clinical relevance.

The applicant should hold a Master's degree (or equivalent, e.g. Engineering degree) in computer vision, data science, computer science or applied mathematics.

Fluency in English, or fluency in French with excellent level in English are mandatory to fill the position.

Software development will primarily be realized in python using PyTorch. A solid practical experience in these language and framework is expected.

• Subsidized meals
• Partial reimbursement of public transport costs
• Leave: 7 weeks of annual leave + 10 extra days off due to RTT (statutory reduction in working hours) + possibility of exceptional leave (sick children, moving home, etc.)
• Possibility of teleworking (after 6 months of employment) and flexible organization of working hours
• Professional equipment available (videoconferencing, loan of computer equipment, etc.)
• Social, cultural and sports events and activities
• Access to vocational training
• Social security coverage

2100€ gross/month the 1st year