The accurate registration of 3D multi-modal medical images taken in different patient positions—such as supine and upright—is of paramount importance for various clinical applications, including diagnosis, treatment planning, and monitoring disease progression. This challenge is particularly critical in scenarios where precise anatomical alignment is necessary for effective treatment implementation. However, position changes and tissue deformations present significant hurdles in achieving accurate image registration. These discrepancies, caused by patient movement and gravity-induced deformations, can obscure crucial anatomical details and impede the alignment required for effective clinical interventions.  

In medical imaging, the task of aligning 3D images taken in different positions is fraught with difficulties due to inherent position changes and tissue deformations. For instance, a patient’s posture shift from  supine to upright can lead to substantial variations in tissue and bone structures, complicating the process of overlaying images accurately. Traditional registration methods often struggle to address these challenges comprehensively, particularly when dealing with non-rigid deformations and anatomical variations. This limitation affects the quality of treatment planning and monitoring, making it imperative to develop more robust and adaptive registration techniques.  

This thesis proposes to develop an innovative end-to-end framework utilizing deep learning techniques to achieve precise registration of 3D multi-modal medical images captured at different patient positions. The primary objectives of this research are:  

1. To develop a deep learning-based model capable of recovering bone structures and articulating models for accurate image alignment. This model will be designed to handle positional changes and align images effectively by incorporating advanced techniques in deep learning. 
2. To integrate a dense deformation model that accounts for tissue and anatomical changes due to gravity. This aspect of the framework will address the complex, non-rigid deformations that occur when a patient shifts from one position to another, ensuring accurate anatomical alignment.
3. To evaluate the developed framework in the context of prostate cancer treatment, specifically focusing on the registration of supine MR scans and upright CT scans for radiotherapy planning. 

The proposed framework will consist of several key components: 

1. Data Collection and Preprocessing: A diverse dataset of 3D multi-modal images will be collected, including supine MR and upright CT scans. These images will be preprocessed to standardize and enhance their quality, ensuring consistent input for the deep learning models. 
2. Deep Learning Model Development: A convolutional neural network (CNN)-based architecture will be designed to handle the registration task. The network will be trained to recover bone structures and articulate models, with a focus on capturing and aligning anatomical details accurately.
3. Dense Deformation Modeling: A sophisticated dense deformation model will be incorporated to  account for the non-rigid deformations due to gravity and positional changes. This model will employ advanced techniques such as deformable image registration and spatial transformations to align images recisely.

The proposed research aims to make several significant contributions:  

1. Development of a novel deep learning-based registration framework that enhances the accuracy of 3D multi-modal image alignment across different patient positions. 
2. Improvement in treatment planning and monitoring through precise image registration, leading to better clinical outcomes and more effective disease management. 
3. Validation of the framework in a clinical setting, demonstrating its practical utility and potential benefits in real-world medical applications.

This thesis seeks to address the pressing challenge of accurate registration of 3D multi-modal medical images taken in different patient positions by developing an advanced deep learning framework. By integrating techniques for bone structure recovery, articulation modeling, and dense deformation handling, the proposed research aims to enhance the precision of image alignment and contribute to improved clinical interventions. The successful implementation and evaluation of this framework in prostate cancer treatment will pave the way for broader applications in various medical fields, ultimately advancing the capabilities of image-guided therapies and diagnostics. 

Applicant profile 

• Higher education in Computer science, Biomedical Engineering, Physics, or a related field.  

• Good knowledge of spoken and written English with academic writing skills. 

• Experience in medical image analysis and deep learning frameworks.  

• Ability to work independently and effectively.  

• Excellent understanding for team working requirements. 

Application process 

If you find this position interesting and would like to work in an exceptional, international, strongly innovative environment, please send your full application documents, including the filled application form, motivation letter, short CV, list of most important publications with explanation of your own contribution, and information of your earliest possible starting date. If you are applying for more than one UPLIFT PhD position, you may indicate your top 1-3 preferences in the application form by using the DC numbers associated with the projects. Please submit your application to [email protected] until January 15, 2025.

We anticipate video conference interviews with candidates starting in the third week of February,
for start dates from March 2025. Applications submitted without any of the required documents will not be considered.  

Candidates can be of any nationality but must not have resided or carried out their main activity (work, studies, etc.) in the country of the recruiting beneficiary for more than 12 months in the 36 months immediately before their recruitment date. Applicants should be within the first four years of their research careers and must not have been awarded a doctoral degree. Submitted applications will be evaluated in accordance with the European Code of Conduct for Recruitment.