R2Net: Efficient and flexible diffeomorphic image registration using Lipschitz continuous residual networks

Radiomics-Guided Deep Learning Networks Classify Differential Diagnosis of Parkinsonism

The differential diagnosis between atypical Parkinsonian syndromes may be challenging and critical. We aimed to proposed a radiomics-guided deep learning (DL) model to discover interpretable DL features and further verify the proposed model through the differential diagnosis of Parkinsonian syndromes. We recruited 1495 subjects for 18F-fluorodeoxyglucose positron emission tomography (18F-FDG PET) scanning, including 220 healthy controls and 1275 patients diagnosed with idiopathic Parkinson's disease (IPD), multiple system atrophy (MSA), or progressive supranuclear palsy (PSP). Baseline radiomics and two DL models were developed and tested for the Parkinsonian diagnosis. The DL latent features were extracted from the last layer and subsequently guided by radiomics. The radiomics-guided DL model outperformed the baseline radiomics approach, suggesting the effectiveness of the DL approach. DenseNet showed the best diagnosis ability (sensitivity: 95.7%, 90.1%, and 91.2% for IPD, MSA, and PSP, respectively) using retained DL features in the test dataset. The retained DL latent features were significantly associated with radiomics features and could be interpreted through biological explanations of handcrafted radiomics features. The radiomics-guided DL model offers interpretable high-level abstract information for differential diagnosis of Parkinsonian disorders and holds considerable promise for personalized disease monitoring.

SMILE: Siamese Multi-scale Interactive-representation LEarning for Hierarchical Diffeomorphic Deformable image registration

Deformable medical image registration plays an important role in many clinical applications. It aims to find a dense deformation field to establish point-wise correspondences between a pair of fixed and moving images. Recently, unsupervised deep learning-based registration methods have drawn more and more attention because of fast inference at testing stage. Despite remarkable progress, existing deep learning-based methods suffer from several limitations including: (a) they often overlook the explicit modeling of feature correspondences due to limited receptive fields; (b) the performance on image pairs with large spatial displacements is still limited since the dense deformation field is regressed from features learned by local convolutions; and (c) desirable properties, including topology-preservation and the invertibility of transformation, are often ignored. To address above limitations, we propose a novel Convolutional Neural Network (CNN) consisting of a Siamese Multi-scale Interactive-representation LEarning (SMILE) encoder and a Hierarchical Diffeomorphic Deformation (HDD) decoder. Specifically, the SMILE encoder aims for effective feature representation learning and spatial correspondence establishing while the HDD decoder seeks to regress the dense deformation field in a coarse-to-fine manner. We additionally propose a novel Local Invertible Loss (LIL) to encourage topology-preservation and local invertibility of the regressed transformation while keeping high registration accuracy. Extensive experiments conducted on two publicly available brain image datasets demonstrate the superiority of our method over the state-of-the-art (SOTA) approaches. Specifically, on the Neurite-OASIS dataset, our method achieved an average DSC of 0.815 and an average ASSD of 0.633 mm.