Evaluation of Learning-based Methods for Multimodal Biomedical Image Registration

• Main Author: Lu, Jiahao
• Format: Others
• Language: English
• Published: Uppsala universitet, Institutionen för informationsteknologi 2020
• Subjects: Engineering and Technology; Teknik och teknologier
• Online Access: http://urn.kb.se/resolve?urn=urn:nbn:se:uu:diva-455890

Multimodal registration of biomedical images, where two or more images are to bemapped into a common coordinate system in order to combine complementaryinformation, is often a highly beneficial yet challenging task. In recent years, the deeplearning renaissance has reactivated the image registration field by showing impressiveperformance in various applications. However, there is still a lack of empiricalevaluations of learning-based methods for registration of multimodal biomedical datain the literature. This study aims to reduce this deficiency by evaluating several promising, whilemethodologically different learning-based registration methods on a dataset consistingof multimodal microscopy images. Selected methods include GAN-basedcross-modality mapping combined with feature- or intensity-based registration, andsupervised or unsupervised end-to-end transformation predictions. Classic iterativemutual information (MI) maximisation and a state-of-the-art framework tunedspecifically to the dataset are used as baselines. Both registration quality andprocessing speed are assessed. In our experiments, none of the learning-based methods surpasses MI maximisation inquality. Nevertheless, GANs are demonstrated useful in extending the ability ofmonomodal registration methods towards multimodal tasks. The outstanding speed ofend-to-end transformation prediction methods in both training and inference stagesmotivates their further exploration. Multi-resolution strategy might be a key toimprove both above-mentioned approaches. The empirical evaluations in this thesisprovide an insight into the challenge in multimodal registration of biomedical images.It not only lays a groundwork that can be used by future research as a reference, butalso points out some promising modifications to be studied further.