Automatic deformable surface registration for medical applications by radial basis function-based robust point-matching

An automatic extraction method on medical feature points based on PointNet++ for robot-assisted knee arthroplasty

BACKGROUND

Image registration is a crucial technology in robot-assisted knee arthroplasty, which provides real-time patient information by registering the pre-operative image data with data acquired during the operation. The existing registration method requires surgeons to manually pick up medical feature points (i.e. anatomical points) in pre-operative images, which is time-consuming and relied on surgeons experience. Moreover, different doctors have different preferences in preoperative planning, which may influence the consistency of surgical results.

METHODS

A medical feature points automatic extraction method based on PointNet++ named Point_RegNet is proposed to improve the efficiency of preoperative preparation and ensure the consistency of surgical results. The proposed method replaces the classification and segmentation layer of PointNet++ with a regression layer to predict the position of feature points. The comparative experiment is adopted to determine the optimal set of abstraction layers in PointNet++.

RESULTS

The proposed network with three set abstraction layers is more suitable for extracting feature points. The feature points predictions mean error of our method is less than 5 mm, which is 1 mm less than the manual marking method. Ultimately, our method only requires less than 3 s to extract all medical feature points in practical application. It is much faster than the manual extraction way which usually requires more than half an hour to mark all necessary feature points.

CONCLUSION

Our deep learning-based method can improve the surgery accuracy and reduce the preoperative preparation time. Moreover, this method can also be applied to other surgical navigation systems.

Developing a three-dimensional statistical shape model of normal dentition using an automated algorithm and normal samples

OBJECTIVES

The statistical shape model (SSM) is a model of geometric properties of a set of shapes based on statistical shape analysis. The SSM develops an average model of several objects using an automated algorithm that excludes the operator's subjectivity. The aim of this study was to develop a three-dimensional (3D) SSM of normal dentition to provide virtual templates for efficient treatment.

MATERIALS AND METHODS

Dental casts were obtained from participants with normal dentition. After acquiring the 3D models, the SSMs of the individual teeth and whole dental arch were generated by an iterative closest point (ICP)-based rigid registration and point correspondences, respectively. Then, the individual tooth SSM was aligned to the whole dental arch SSM using ICP-based registration to generate an average model of normal dentition.

RESULTS

The generated 3D SSM showed specific morphological features of normal dentition similar to those previously reported. Moreover, on measuring the arch dimensions, all values in this study were similar to those previously reported using normal dentition.

CONCLUSIONS

The 3D SSM of normal dentition may increase the diagnostic efficiency of orthodontic treatments by providing a visual objective. It can be also used as a 3D template in various fields of dentistry.

CLINICAL RELEVANCE

Our SSM of normal dentition provides both quantitative and qualitative information on the 3D morphology of teeth and dental arches, which may provide valuable information on 3D virtual-setup, bracket fabrication, and aligner treatment.

Biomechanically constrained non-rigid MR-TRUS prostate registration using deep learning based 3D point cloud matching

A non-rigid MR-TRUS image registration framework is proposed for prostate interventions. The registration framework consists of a convolutional neural networks (CNN) for MR prostate segmentation, a CNN for TRUS prostate segmentation and a point-cloud based network for rapid 3D point cloud matching. Volumetric prostate point clouds were generated from the segmented prostate masks using tetrahedron meshing. The point cloud matching network was trained using deformation field that was generated by finite element analysis. Therefore, the network implicitly models the underlying biomechanical constraint when performing point cloud matching. A total of 50 patients' datasets were used for the network training and testing. Alignment of prostate shapes after registration was evaluated using three metrics including Dice similarity coefficient (DSC), mean surface distance (MSD) and Hausdorff distance (HD). Internal point-to-point registration accuracy was assessed using target registration error (TRE). Jacobian determinant and strain tensors of the predicted deformation field were calculated to analyze the physical fidelity of the deformation field. On average, the mean and standard deviation were 0.94±0.02, 0.90±0.23 mm, 2.96±1.00 mm and 1.57±0.77 mm for DSC, MSD, HD and TRE, respectively. Robustness of our method to point cloud noise was evaluated by adding different levels of noise to the query point clouds. Our results demonstrated that the proposed method could rapidly perform MR-TRUS image registration with good registration accuracy and robustness.

Using machine learning methods to determine a typology of patients with HIV-HCV infection to be treated with antivirals

Several European countries have established criteria for prioritising initiation of treatment in patients infected with the hepatitis C virus (HCV) by grouping patients according to clinical characteristics. Based on neural network techniques, our objective was to identify those factors for HIV/HCV co-infected patients (to which clinicians have given careful consideration before treatment uptake) that have not being included among the prioritisation criteria. This study was based on the Spanish HERACLES cohort (NCT02511496) (April-September 2015, 2940 patients) and involved application of different neural network models with different basis functions (product-unit, sigmoid unit and radial basis function neural networks) for automatic classification of patients for treatment. An evolutionary algorithm was used to determine the architecture and estimate the coefficients of the model. This machine learning methodology found that radial basis neural networks provided a very simple model in terms of the number of patient characteristics to be considered by the classifier (in this case, six), returning a good overall classification accuracy of 0.767 and a minimum sensitivity (for the classification of the minority class, untreated patients) of 0.550. Finally, the area under the ROC curve was 0.802, which proved to be exceptional. The parsimony of the model makes it especially attractive, using just eight connections. The independent variable “recent PWID” is compulsory due to its importance. The simplicity of the model means that it is possible to analyse the relationship between patient characteristics and the probability of belonging to the treated group.

Anisotropic non-rigid Iterative Closest Point Algorithm for respiratory motion abdominal surface matching

Surface registration is a one of the crucial and actual problems of computer aided surgery. This paper presents the modification of the non-rigid Iterative Closest Point Algorithm which takes into account an anisotropic noise model and landmarks as guided correspondence at the transformation step in every iteration. The presented approach was validated on human abdominal briefing surface data from a time-of-flight camera. We took the median of the resulting measures and the outcome is presented: the median of means of surfaces distance was at the same level for both variants of the ICP algorithm and is comparable with the isotropic variant, the median of mean landmark position errors decreased by 0.93 units (over 20% improvement) and the median of percentage of single correspondences in target point cloud increased by 11.96%. The results showed that the introduction of the anisotropic model of noise for the ToF camera allows for the improvement the percentage of target cloud points which had only one correspondent over 10% impartment and additional weighting of markers also improves the measure of the quality of finding real correspondents over 20% improvement. In the examined dataset, where the average initial distance between the clouds of points in the inspiratory and expiration is equal to approx. 7.5 mm, a more than 10% improvement in the quality of the correspondence improves the accuracy of matching the surface within 1 mm which is a significant value in application of minimally invasive image guided interventions.