Learning curve for fusion magnetic resonance imaging targeted prostate biopsy and three-dimensional transrectal ultrasonography segmentation.

OBJECTIVE: To report the learning curve of multiple operators for fusion magnetic resonance imaging (MRI) targeted biopsy and to determine the number of cases needed to achieve proficiency. MATERIALS AND METHODS: All adult males who underwent fusion MRI targeted biopsy between February 2012 and July 2021 for clinically suspected prostate cancer (PCa) in a single centre were included. Fusion transrectal MRI targeted biopsy was performed under local anaesthesia using the Koelis platform. Learning curves for segmentation of transrectal ultrasonography (TRUS) images and the overall MRI targeted biopsy procedure were estimated with locally weighted scatterplot smoothing by computing each operator's timestamps for consecutive procedures. Non-risk-adjusted cumulative sum (CUSUM) methods were used to create learning curves for clinically significant (i.e., International Society of Urological Pathology grade ≥ 2) PCa detection. RESULTS: Overall, 1721 patients underwent MRI targeted biopsy in our centre during the study period. The median (interquartile range) times for TRUS segmentation and for the MRI targeted biopsy procedure were 4.5 (3.5, 6.0) min and 13.2 (10.6, 16.9) min, respectively. Among the 14 operators with experience of more than 50 cases, a plateau was reached after 40 cases for TRUS segmentation time and 50 cases for overall MRI targeted biopsy procedure time. CUSUM analysis showed that the learning curve for clinically significant PCa detection required 25 to 45 procedures to achieve clinical proficiency. Pain scores ranged between 0 and 1 for 84% of patients, and a plateau phase was reached after 20 to 100 cases. CONCLUSIONS: A minimum of 50 cases of MRI targeted biopsy are necessary to achieve clinical and technical proficiency and to reach reproducibility in terms of timing, clinically significant PCa detection, and pain.

Impact of Relative Volume Difference Between Magnetic Resonance Imaging and Three-dimensional Transrectal Ultrasound Segmentation on Clinically Significant Prostate Cancer Detection in Fusion Magnetic Resonance Imaging-targeted Biopsy.

BACKGROUND: Segmentation of three-dimensional (3D) transrectal ultrasound (TRUS) images is known to be challenging, and the clinician often lacks a reliable and easy-to-use indicator to assess its accuracy during the fusion magnetic resonance imaging (MRI)-targeted prostate biopsy procedure. OBJECTIVE: To assess the effect of the relative volume difference between 3D-TRUS and MRI segmentation on the outcome of a targeted biopsy. DESIGN, SETTING, AND PARTICIPANTS: All adult males who underwent an MRI-targeted prostate biopsy for clinically suspected prostate cancer between February 2012 and July 2021 were consecutively included. INTERVENTION: All patients underwent a fusion MRI-targeted prostate biopsy with a Koelis device. OUTCOME MEASUREMENTS AND STATISTICAL ANALYSIS: Three-dimensional TRUS and MRI prostate volumes were calculated using 3D prostate models issued from the segmentations. The primary outcome was the relative segmentation volume difference (SVD) between transrectal ultrasound and MRI divided by the MRI volume (SVD = MRI volume - TRUS volume/MRI volume) and its correlation with clinically significant prostate cancer (eg, International Society of Urological Pathology [ISUP] ≥2) positiveness on targeted biopsy cores. RESULTS AND LIMITATIONS: Overall, 1721 patients underwent a targeted biopsy resulting in a total of 5593 targeted cores. The median relative SVD was significantly lower in patients diagnosed with clinically significant prostate cancer than in those with ISUP 0-1: (6.7% [interquartile range {IQR} -2.7, 13.6] vs 8.0% [IQR 3.3, 16.4], p < 0.01). A multivariate regression analysis showed that a relative SVD of >10% of the MRI volume was associated with a lower detection rate of clinically significant prostate cancer (odds ratio = 0.74 [95% confidence interval: 0.55-0.98]; p = 0.038). CONCLUSIONS: A relative SVD of >10% of the MRI segmented volume was associated with a lower detection rate of clinically significant prostate cancer on targeted biopsy cores. The relative SVD can be used as a per-procedure quality indicator of 3D-TRUS segmentation. PATIENT SUMMARY: A discrepancy of ≥10% between segmented magnetic resonance imaging and transrectal ultrasound volume is associated with a reduced ability to detect significant prostate cancer on targeted biopsy cores.

2D/3D Deep Registration Along Trajectories With Spatiotemporal Context: Application to Prostate Biopsy Navigation.

OBJECTIVE: The accuracy of biopsy targeting is a major issue for prostate cancer diagnosis and therapy. However, navigation to biopsy targets remains challenging due to the limitations of transrectal ultrasound (TRUS) guidance added to prostate motion issues. This article describes a rigid 2D/3D deep registration method, which provides a continuous tracking of the biopsy location w.r.t the prostate for enhanced navigation. METHODS: A spatiotemporal registration network (SpT-Net) is proposed to localize the live 2D US image relatively to a previously aquired US reference volume. The temporal context relies on prior trajectory information based on previous registration results and probe tracking. Different forms of spatial context were compared through inputs (local, partial or global) or using an additional spatial penalty term. The proposed 3D CNN architecture with all combinations of spatial and temporal context was evaluated in an ablation study. For providing a realistic clinical validation, a cumulative error was computed through series of registrations along trajectories, simulating a complete clinical navigation procedure. We also proposed two dataset generation processes with increasing levels of registration complexity and clinical realism. RESULTS: The experiments show that a model using local spatial information combined with temporal information performs better than more complex spatiotemporal combination. CONCLUSION: The best proposed model demonstrates robust real-time 2D/3D US cumulated registration performance on trajectories. Those results respect clinical requirements, application feasibility, and they outperform similar state-of-the-art methods. SIGNIFICANCE: Our approach seems promising for clinical prostate biopsy navigation assistance or other US image-guided procedure.

Multi-eXpert fusion: An ensemble learning framework to segment 3D TRUS prostate images.

PURPOSE: Prostate segmentation of 3D TRUS images is a prerequisite for several diagnostic and therapeutic applications. Unfortunately, this difficult task suffers from high intra and interobserver variability, even for experienced urologists/radiologists. This is why automatic segmentation algorithms could have a significant clinical added-value. METHODS: This paper introduces a new deep segmentation architecture consisting of two main phases: view-specific segmentations of 2D slices and their fusion. The segmentation phase is based on three segmentation networks trained in parallel on specific slice viewing directions: axial, coronal, and sagittal. The proposed fusion network is then fed with the output of the segmentation networks and trained to produce three confidence maps. These maps correspond to the local trust granted by the fusion network to each view-specific segmentation network. Finally, for a given slice, the segmentation is computed by combining these confidence maps with their corresponding segmentations. The 3D segmentation of the prostate is obtained by restacking all the segmented slices to form a volume. RESULTS: This approach was evaluated on a database of 100 patients with several combinations of network architectures (for both the segmentation phase and the fusion phase) to show the flexibility and reliability of the framework. The proposed approach was also compared to STAPLE, to the majority voting strategy, and to a direct 3D approach tested on the same database. The new method outperforms these three approaches on all evaluation criteria. Finally, the results of the multi-eXpert fusion (MXF) framework compare favorably with other state-of-the-art methods, while these methods typically work on smaller databases. CONCLUSIONS: We proposed a novel MXF framework to segment 3D TRUS images of the prostate. The main feature of this approach is the fusion of expert network results at the pixel level using computed confidence maps. Experiments conducted on a clinical database have shown the robustness and flexibility of this approach and its superiority over state-of-the-art approaches. Finally, the MXF framework demonstrated its ability to capture and preserve the underlying gland structures, particularly in the base and apex regions.

Towards real-time free-hand biopsy navigation.

PURPOSE: Performing a transrectal ultrasound (TRUS) prostate biopsy is at the heart of the current prostate cancer detection procedure. With today's two-dimensional (2D) live ultrasound (US) imaging equipment, this task remains complex due to the poor visibility of cancerous tissue on TRUS images and the limited anatomical context available in the 2D TRUS plane. This paper presents a rigid 2D/3DUS registration method for navigated prostate biopsy. This allows continuous localization of the biopsy trajectory during the procedure. METHODS: We proposed an organ-based approach to achieve real-time rigid registration without the need for any probe localization device. The registration method combines image similarity and geometric proximity of detected features. Additions to our previous work include a multi-level approach and the use of a rejection rate favouring the best matches. Their aim is to increase the accuracy and time performances. These modifications and their in-depth evaluation on real clinical cases and comparison to this previous work are described. We performed static and dynamic evaluations along biopsy trajectories on a very large amount of data acquired under uncontrolled routine conditions. The computed transforms are compared to a ground truth obtained either from corresponding manually detected fiducials or from an already evaluated registration method. RESULTS: All results show that the current method outperforms its previous version, both in terms of accuracy (the average error reported here is 12 to 17% smaller depending on the experiment) and processing time (from 20 to 60 times faster compared to the previous implementation). The dynamic registration experiment demonstrates that the method can be successfully used for continuous tracking of the biopsy location w.r.t the prostate at a rate that varies between 5 and 15 Hz. CONCLUSIONS: This work shows that on the fly 2D/3DUS registration can be performed very efficiently on biopsy trajectories. This allows us to plan further improvements in prostate navigation and a clinical transfer.

Mutual cineMR/RT3DUS cardiac segmentation.

We present a new method to segment a cardiac RT3D ultrasound volume by integrating the registered segmentation of a cardiac cine-MR series in short axis of the same patient. The motivation behind our method is to improve the ultrasound segmentation process by integrating a reference shape built using the cine-MR segmentation on the same patient. As a side effect we obtain a close registration of the cine MR short axis slices with respect to the ultrasound volume. We use the level set framework with a functional including a region-based and a shape-based term. The reference shape is iteratively registered onto the contour during the ultrasound segmentation process and using an affine transform. The proposed method is demonstrated on the MICCAI11 Motion Tracking Challenge database.

Using infrared thermography to study hydrostatic stress networks in granular materials.

The macroscopic mechanical behaviour of granular materials is governed by microscopic features at the particle scale. Photoelasticimetry is a powerful method for measuring shear stresses in particles made from birefringent materials. As a complementary method, we here identify the hydrostatic stress networks through thermoelastic stress analysis using infrared thermographic measurements. Experiments are performed on two-dimensional cohesionless monodisperse granular materials composed of about 1200 cylinders comprising two constitutive materials. We show that the experimental hydrostatic stress distributions follow statistical laws which are in agreement with simulations performed using molecular dynamics, except in one case exhibiting piecewise periodic stacking. Polydisperse cases are then processed. The measurement of hydrostatic stress networks using this technique opens new prospects for the analysis of granular materials.