Can representation learning for multimodal image registration be improved by supervision of intermediate layers?

Multimodal imaging and correlative analysis typically require image alignment. Contrastive learning can generate representations of multimodal images, reducing the challenging task of multimodal image registration to a monomodal one. Previously, additional supervision on intermediate layers in contrastive learning has improved biomedical image classification. We evaluate if a similar approach improves representations learned for registration to boost registration performance. We explore three approaches to add contrastive supervision to the latent features of the bottleneck layer in the U-Nets encoding the multimodal images and evaluate three different critic functions. Our results show that representations learned without additional supervision on latent features perform best in the downstream task of registration on two public biomedical datasets. We investigate the performance drop by exploiting recent insights in contrastive learning in classification and self-supervised learning. We visualize the spatial relations of the learned representations by means of multidimensional scaling, and show that additional supervision on the bottleneck layer can lead to partial dimensional collapse of the intermediate embedding space.

Cross-Modality Sub-Image Retrieval using Contrastive Multimodal Image Representations

In tissue characterization and cancer diagnostics, multimodal imaging has emerged as a powerful technique. Thanks to computational advances, large datasets can be exploited to improve diagnosis and discover patterns in pathologies. However, this requires efficient and scalable image retrieval methods. Cross-modality image retrieval is particularly demanding, as images of the same content captured in different modalities may display little common information. We propose a content-based image retrieval system (CBIR) for reverse (sub-)image search to retrieve microscopy images in one modality given a corresponding image captured by a different modality, where images are not aligned and share only few structures. We propose to combine deep learning to generate representations which embed both modalities in a common space, with classic, fast, and robust feature extractors (SIFT, SURF) to create a bag-of-words model for efficient and reliable retrieval. Our application-independent approach shows promising results on a publicly available dataset of brightfield and second harmonic generation microscopy images. We obtain 75.4% and 83.6% top-10 retrieval success for retrieval in one or the other direction. Our proposed method significantly outperforms both direct retrieval of the original multimodal (sub-)images, as well as their corresponding generative adversarial network (GAN)-based image-to-image translations. We establish that the proposed method performs better in comparison with a recent sub-image retrieval toolkit, GAN-based image-to-image translations, and learnt feature extractors for the downstream task of cross-modal image retrieval. We highlight the shortcomings of the latter methods and observe the importance of equivariance and invariance properties of the learnt representations and feature extractors in the CBIR pipeline. Code will be available at github.com/MIDA-group.

CoMIR: Contrastive Multimodal Image Representation for Registration

We propose contrastive coding to learn shared, dense image representations, referred to as CoMIRs (Contrastive Multimodal Image Representations). CoMIRs enable the registration of multimodal images where existing registration methods often fail due to a lack of sufficiently similar image structures. CoMIRs reduce the multimodal registration problem to a monomodal one in which general intensity-based, as well as feature-based, registration algorithms can be applied. The method involves training one neural network per modality on aligned images, using a contrastive loss based on noise-contrastive estimation (InfoNCE). Unlike other contrastive coding methods, used for e.g. classification, our approach generates image-like representations that contain the information shared between modalities. We introduce a novel, hyperparameter-free modification to InfoNCE, to enforce rotational equivariance of the learnt representations, a property essential to the registration task. We assess the extent of achieved rotational equivariance and the stability of the representations with respect to weight initialization, training set, and hyperparameter settings, on a remote sensing dataset of RGB and near-infrared images. We evaluate the learnt representations through registration of a biomedical dataset of bright-field and second-harmonic generation microscopy images; two modalities with very little apparent correlation. The proposed approach based on CoMIRs significantly outperforms registration of representations created by GAN-based image-to-image translation, as well as a state-of-the-art, application-specific method which takes additional knowledge about the data into account. Code is available at: https://github.com/dqiamsdoayehccdvulyy/CoMIR.