Harnessing clinical annotations to improve deep learning performance in prostate segmentation

Purpose Developing large-scale datasets with research-quality annotations is challenging due to the high cost of refining clinically generated markup into high precision annotations. We evaluated the direct use of a large dataset with only clinically generated annotations in development of high-performance segmentation models for small research-quality challenge datasets. Materials and methods We used a large retrospective dataset from our institution comprised of 1,620 clinically generated segmentations, and two challenge datasets (PROMISE12: 50 patients, ProstateX-2: 99 patients). We trained a 3D U-Net convolutional neural network (CNN) segmentation model using our entire dataset, and used that model as a template to train models on the challenge datasets. We also trained versions of the template model using ablated proportions of our dataset, and evaluated the relative benefit of those templates for the final models. Finally, we trained a version of the template model using an out-of-domain brain cancer dataset, and evaluated the relevant benefit of that template for the final models. We used five-fold cross-validation (CV) for all training and evaluation across our entire dataset. Results Our model achieves state-of-the-art performance on our large dataset (mean overall Dice 0.916, average Hausdorff distance 0.135 across CV folds). Using this model as a pre-trained template for refining on two external datasets significantly enhanced performance (30% and 49% enhancement in Dice scores respectively). Mean overall Dice and mean average Hausdorff distance were 0.912 and 0.15 for the ProstateX-2 dataset, and 0.852 and 0.581 for the PROMISE12 dataset. Using even small quantities of data to train the template enhanced performance, with significant improvements using 5% or more of the data. Conclusion We trained a state-of-the-art model using unrefined clinical prostate annotations and found that its use as a template model significantly improved performance in other prostate segmentation tasks, even when trained with only 5% of the original dataset.

Calibration of a Heterogeneous Brain Model Using a Subject-Specific Inverse Finite Element Approach

Central to the investigation of the biomechanics of traumatic brain injury (TBI) and the assessment of injury risk from head impact are finite element (FE) models of the human brain. However, many existing FE human brain models have been developed with simplified representations of the parenchyma, which may limit their applicability as an injury prediction tool. Recent advances in neuroimaging techniques and brain biomechanics provide new and necessary experimental data that can improve the biofidelity of FE brain models. In this study, the CAB-20MSym template model was developed, calibrated, and extensively verified. To implement material heterogeneity, a magnetic resonance elastography (MRE) template image was leveraged to define the relative stiffness gradient of the brain model. A multi-stage inverse FE (iFE) approach was used to calibrate the material parameters that defined the underlying non-linear deviatoric response by minimizing the error between model-predicted brain displacements and experimental displacement data. This process involved calibrating the infinitesimal shear modulus of the material using low-severity, low-deformation impact cases and the material non-linearity using high-severity, high-deformation cases from a dataset of in situ brain displacements obtained from cadaveric specimens. To minimize the geometric discrepancy between the FE models used in the iFE calibration and the cadaveric specimens from which the experimental data were obtained, subject-specific models of these cadaveric brain specimens were developed and used in the calibration process. Finally, the calibrated material parameters were extensively verified using independent brain displacement data from 33 rotational head impacts, spanning multiple loading directions (sagittal, coronal, axial), magnitudes (20–40 rad/s), durations (30–60 ms), and severity. Overall, the heterogeneous CAB-20MSym template model demonstrated good biofidelity with a mean overall CORA score of 0.63 ± 0.06 when compared to in situ brain displacement data. Strains predicted by the calibrated model under non-injurious rotational impacts in human volunteers (N = 6) also demonstrated similar biofidelity compared to in vivo measurements obtained from tagged magnetic resonance imaging studies. In addition to serving as an anatomically accurate model for further investigations of TBI biomechanics, the MRE-based framework for implementing material heterogeneity could serve as a foundation for incorporating subject-specific material properties in future models.

Modelling, Control and Design of a Clutched Parallel Elastically Actuated Articulated Robotic Leg Through Virtual Tunable Damping

In this study, design, modelling and control of a clutched parallel elastically actuated articulated leg is presented. Clutch mechanism is introduced to disengage the parallel elastic element when it is not needed. Some of the design principles concerning the ease of manufacturing and assembly are underlined. While the system has two joints at hip and knee that can be actuated, for simplicity, restrained motion of the system in vertical direction is considered only with hip actuation. Controller is based on a template model and the desired motion is obtained by equating (embedding) dynamics of the physical system (anchor) to the template model. Spring loaded inverted pendulum (SLIP) model including a virtual viscous damper is chosen as the template. Controller decides on the virtual damping constant in the template to reach desired apex positions. A wrapping cam mechanism is introduced to equate the potential energy function of the parallel spring to the desired linear spring of SLIP model. To complete embedding, necessary torque is calculated by equating the virtual works of the inputs. Overall, simulation of the hopping system and the important aspects of design are presented.

Elaboration of the measuring procedure facilitating precision assessment of the geometry of mandible anatomical model manufactured using additive methods

This article presents a procedure for minimizing ATOS II Triple Scan system measurement errors during the verification of geometrical accuracy of the final lateral-mandibular condyle model. The process of manufacturing a template geometrically similar to that of lateral-mandibular condyle was performed on the five-axis machining centre 100 DMU MonoBlock. The next stage of the research was related to the implementation of the measurement system procedure on the template model, and the 12 anatomical models of the mandibular body-condyle were manufactured using five different additive methods. As a result of the comparison of anatomical models of the mandibular body-condyle designed in reverse engineering/computer-aided design systems and manufactured using additive methods, the average results of histograms and parameters determining the accuracy of geometry of 12 models were obtained. In the case of models manufactured using fused deposition modelling, PolyJet and selective laser sintering techniques, a unimodal distribution was observed in the same way as in the template model. The best results were obtained in the case of models manufactured using selective laser sintering techniques (standard deviation = 0.06 mm). In the case of fused deposition modelling and PolyJet, a similar value of standard deviation (about 0.07 mm) was observed, despite the fact that the layer thickness for PolyJet technology was 0.016 mm. In the case of melted and extruded modelling and ColorJet Printing technologies, there was a bimodal distribution. Through the implementation of own template and measurement method, it will be easier to estimate errors in the manufacturing of anatomical models of lateral-mandibular condyle part. As a result, medical models, surgical templates and implants will be manufactured more accurately and precisely, which will significantly reduce intraoperative complications during the surgical procedure in this area.