Toward confident prostate cancer detection using ultrasound: a multi-center study.

PURPOSE: Deep learning-based analysis of micro-ultrasound images to detect cancerous lesions is a promising tool for improving prostate cancer (PCa) diagnosis. An ideal model should confidently identify cancer while responding with appropriate uncertainty when presented with out-of-distribution inputs that arise during deployment due to imaging artifacts and the biological heterogeneity of patients and prostatic tissue. METHODS: Using micro-ultrasound data from 693 patients across 5 clinical centers who underwent micro-ultrasound guided prostate biopsy, we train and evaluate convolutional neural network models for PCa detection. To improve robustness to out-of-distribution inputs, we employ and comprehensively benchmark several state-of-the-art uncertainty estimation methods. RESULTS: PCa detection models achieve performance scores up to 76 % average AUROC with a 10-fold cross validation setup. Models with uncertainty estimation obtain expected calibration error scores as low as 2 % , indicating that confident predictions are very likely to be correct. Visualizations of the model output demonstrate that the model correctly identifies healthy versus malignant tissue. CONCLUSION: Deep learning models have been developed to confidently detect PCa lesions from micro-ultrasound. The performance of these models, determined from a large and diverse dataset, is competitive with visual analysis of magnetic resonance imaging, the clinical benchmark to identify PCa lesions for targeted biopsy. Deep learning with micro-ultrasound should be further studied as an avenue for targeted prostate biopsy.

Adaptation of a Clinical High-Frequency Transrectal Ultrasound System for Prostate Photoacoustic Imaging: Implementation and Pre-clinical Demonstration.

OBJECTIVE: High-frequency, high-resolution transrectal micro-ultrasound (micro-US: ≥15 MHz) imaging of the prostate is emerging as a beneficial tool for scoring disease risk and accurately targeting biopsies. Adding photoacoustic (PA) imaging to visualize abnormal vascularization and accumulation of contrast agents in tumors has potential for guiding focal therapies. In this work, we describe a new imaging platform that combines a transrectal micro-US system with transurethral light delivery for PA imaging. METHODS: A clinical transrectal micro-US system was adapted to acquire PA images synchronous to a tunable laser pulse. A transurethral side-firing optical fiber was developed for light delivery. A polyvinyl chloride (PVC)-plastisol phantom was developed and characterized to image PA contrast agents in wall-less channels. After resolution measurement in water, PA imaging was demonstrated in phantom channels with dyes and biodegradable nanoparticle contrast agents called porphysomes. In vivo imaging of a tumor model was performed, with porphysomes administered intravenously. RESULTS: Photoacoustic imaging data were acquired at 5 Hz, and image reconstruction was performed offline. PA image resolution at a 14-mm depth was 74 and 261 µm in the axial and lateral directions, respectively. The speed of sound in PVC-plastisol was 1383 m/s, and the attenuation was 4 dB/mm at 20 MHz. PA signal from porphysomes was spectrally unmixed from blood signals in the tumor, and a signal increase was observed 3 h after porphysome injection. CONCLUSION: A combined transrectal micro-US and PA imaging system was developed and characterized, and in vivo imaging demonstrated. High-resolution PA imaging may provide valuable additional information for diagnostic and therapeutic applications in the prostate.

Self-Supervised Learning With Limited Labeled Data for Prostate Cancer Detection in High-Frequency Ultrasound.

Deep learning-based analysis of high-frequency, high-resolution micro-ultrasound data shows great promise for prostate cancer (PCa) detection. Previous approaches to analysis of ultrasound data largely follow a supervised learning (SL) paradigm. Ground truth labels for ultrasound images used for training deep networks often include coarse annotations generated from the histopathological analysis of tissue samples obtained via biopsy. This creates inherent limitations on the availability and quality of labeled data, posing major challenges to the success of SL methods. However, unlabeled prostate ultrasound data are more abundant. In this work, we successfully apply self-supervised representation learning to micro-ultrasound data. Using ultrasound data from 1028 biopsy cores of 391 subjects obtained in two clinical centers, we demonstrate that feature representations learned with this method can be used to classify cancer from noncancer tissue, obtaining an AUROC score of 91% on an independent test set. To the best of our knowledge, this is the first successful end-to-end self-SL (SSL) approach for PCa detection using ultrasound data. Our method outperforms baseline SL approaches, generalizes well between different data centers, and scales well in performance as more unlabeled data are added, making it a promising approach for future research using large volumes of unlabeled data. Our code is publicly available at https://www.github.com/MahdiGilany/SSL_micro_ultrasound.

TRUSformer: improving prostate cancer detection from micro-ultrasound using attention and self-supervision.

PURPOSE: A large body of previous machine learning methods for ultrasound-based prostate cancer detection classify small regions of interest (ROIs) of ultrasound signals that lie within a larger needle trace corresponding to a prostate tissue biopsy (called biopsy core). These ROI-scale models suffer from weak labeling as histopathology results available for biopsy cores only approximate the distribution of cancer in the ROIs. ROI-scale models do not take advantage of contextual information that are normally considered by pathologists, i.e., they do not consider information about surrounding tissue and larger-scale trends when identifying cancer. We aim to improve cancer detection by taking a multi-scale, i.e., ROI-scale and biopsy core-scale, approach. METHODS: Our multi-scale approach combines (i) an "ROI-scale" model trained using self-supervised learning to extract features from small ROIs and (ii) a "core-scale" transformer model that processes a collection of extracted features from multiple ROIs in the needle trace region to predict the tissue type of the corresponding core. Attention maps, as a by-product, allow us to localize cancer at the ROI scale. RESULTS: We analyze this method using a dataset of micro-ultrasound acquired from 578 patients who underwent prostate biopsy, and compare our model to baseline models and other large-scale studies in the literature. Our model shows consistent and substantial performance improvements compared to ROI-scale-only models. It achieves [Formula: see text] AUROC, a statistically significant improvement over ROI-scale classification. We also compare our method to large studies on prostate cancer detection, using other imaging modalities. CONCLUSIONS: Taking a multi-scale approach that leverages contextual information improves prostate cancer detection compared to ROI-scale-only models. The proposed model achieves a statistically significant improvement in performance and outperforms other large-scale studies in the literature. Our code is publicly available at www.github.com/med-i-lab/TRUSFormer .

Robotically controlled three-dimensional micro-ultrasound for prostate biopsy guidance.

PURPOSE: Prostate imaging to guide biopsy remains unsatisfactory, with current solutions suffering from high complexity and poor accuracy and reliability. One novel entrant into this field is micro-ultrasound (microUS), which uses a high-frequency imaging probe to achieve very high spatial resolution, and achieves prostate cancer detection rates equivalent to multiparametric magnetic resonance imaging (mpMRI). However, the ExactVu transrectal microUS probe has a unique geometry that makes it challenging to acquire controlled, repeatable three-dimensional (3D) transrectal ultrasound (TRUS) volumes. We describe the design, fabrication, and validation of a 3D acquisition system that allows for the accurate use of the ExactVu microUS device for volumetric prostate imaging. METHODS: The design uses a motorized, computer-controlled brachytherapy stepper to rotate the ExactVu transducer about its axis. We perform geometric validation using a phantom with known dimensions and compare performance with magnetic resonance imaging (MRI) using a commercial quality assurance anthropomorphic prostate phantom. RESULTS: Our geometric validation shows accuracy of 1 mm or less in all three directions, and images of an anthropomorphic phantom qualitatively match those acquired using MRI and show good agreement quantitatively. CONCLUSION: We describe the first system to acquire robotically controlled 3D microUS images using the ExactVu microUS system. The reconstructed 3D microUS images are accurate, which will allow for future applications of the ExactVu microUS system in prostate specimen and in vivo imaging.

Improving Prostate Cancer (PCa) Classification Performance by Using Three-Player Minimax Game to Reduce Data Source Heterogeneity.

PCa is a disease with a wide range of tissue patterns and this adds to its classification difficulty. Moreover, the data source heterogeneity, i.e. inconsistent data collected using different machines, under different conditions, by different operators, from patients of different ethnic groups, etc., further hinders the effectiveness of training a generalized PCa classifier. In this paper, for the first time, a Generative Adversarial Network (GAN)-based three-player minimax game framework is used to tackle data source heterogeneity and to improve PCa classification performance, where a proposed modified U-Net is used as the encoder. Our dataset consists of novel high-frequency ExactVu ultrasound (US) data collected from 693 patients at five data centers. Gleason Scores (GSs) are assigned to the 12 prostatic regions of each patient. Two classification tasks: benign vs. malignant and low- vs. high-grade, are conducted and the classification results of different prostatic regions are compared. For benign vs. malignant classification, the three-player minimax game framework achieves an Area Under the Receiver Operating Characteristic (AUC) of 93.4%, a sensitivity of 95.1% and a specificity of 87.7%, respectively, representing significant improvements of 5.0%, 3.9%, and 6.0% compared to those of using heterogeneous data, which confirms its effectiveness in terms of PCa classification.

Comparison of conventional transrectal ultrasound, magnetic resonance imaging, and micro-ultrasound for visualizing prostate cancer in an active surveillance population: A feasibility study.

INTRODUCTION: Active surveillance monitoring of prostate cancer is unique in that most patients have low-grade disease that is not well-visualized by any common imaging technique. High-resolution (29 MHz) micro-ultrasound is a new, real-time modality that has been demonstrated to be sensitive to significant prostate cancer and effective for biopsy targeting. This study compares micro-ultrasound imaging with magnetic resonance imaging (MRI) and conventional ultrasound for visualizing prostate cancer in active surveillance. METHODS: Nine patients on active surveillance were imaged with multiparametric (mp) MRI prior to biopsy. During the biopsy procedure, imaging and target identification was first performed using conventional ultrasound, then using micro-ultrasound. The mpMRI report was then unblinded and used to determine cognitive fusion targets. Using micro-ultrasound, biopsy samples were taken from targets in each modality, plus 12 systematic samples. RESULTS: mpMRI and micro-ultrasound both demonstrated superior sensitivity to Gleason sum 7 or higher cancer compared to conventional ultrasound (p=0.02 McNemar's test). Micro-ultrasound detected 89% of clinically significant cancer, compared to 56% for mpMRI. CONCLUSIONS: Micro-ultrasound may provide similar sensitivity to clinically significant prostate cancer as mpMRI and visualize all significant mpMRI targets. Unlike mpMRI, micro-ultrasound is performed in the office, in real-time during the biopsy procedure, and so is expected to maintain the cost-effectiveness of conventional ultrasound. Larger studies are needed before these results may be applied in a clinical setting.

High-Frequency Quantitative Ultrasound for Imaging Prostate Cancer Using a Novel Micro-Ultrasound Scanner.

Currently, biopsies guided by transrectal ultrasound (TRUS) are the only method for definitive diagnosis of prostate cancer. Studies by our group suggest that quantitative ultrasound (QUS) could provide a more sensitive means of targeting biopsies and directing focal treatments to cancer-suspicious regions in the prostate. Previous studies have utilized ultrasound signals at typical clinical frequencies, i.e., in the 6-MHz range. In the present study, a 29-MHz, TRUS, micro-ultrasound system and transducer (ExactVu micro-ultrasound, Exact Imaging, Markham, Canada) was used to acquire radio frequency data from 163 patients immediately before 12-core biopsy procedures, comprising 1956 cores. These retrospective data are a subset of data acquired in an ongoing, multisite, 2000-patient, randomized, clinical trial (clinicaltrials.gov NCT02079025). Spectrum-based QUS estimates of effective scatter diameter (ESD), effective acoustic concentration (EAC), midband (M), intercept (I) and slope (S) as well as envelope statistics employing a Nakagami distribution were used to train linear discriminant classifiers (LDCs) and support vector machines (SVMs). Classifier performance was assessed using area-under-the-curve (AUC) values obtained from receiver operating characteristic (ROC) analyses with 10-fold cross validation. A combination of ESD and EAC parameters resulted in an AUC value of 0.77 using a LDC. When Nakagami-µ or prostate-specific antigen (PSA) values were added as features, the AUC value increased to 0.79. SVM produced an AUC value of 0.77, using a combination of envelope and spectral QUS estimates. The best classification produced an AUC value of 0.81 using an LDC when combining envelope statistics, PSA, ESD and EAC. In a previous study, B-mode-based scoring and evaluation using the PRI-MUS protocol produced a maximal AUC value of 0.74 for higher Gleason-score values (GS >7) when read by an expert. Our initial results with AUC values of 0.81 are very encouraging for developing a new, predominantly user-independent, prostate-cancer, risk-assessing tool.

Assessing Cancer Risk on Novel 29 MHz Micro-Ultrasound Images of the Prostate: Creation of the Micro-Ultrasound Protocol for Prostate Risk Identification.

PURPOSE: Conventional ultrasound systems operate at 6 to 9 MHz and serve as the standard of care to guide prostate biopsies. We present a protocol using a novel high resolution (29 MHz) transrectal prostate micro-ultrasound system. This protocol includes a scoring system to assess the risk of prostatic carcinoma and enable real-time targeted biopsies. MATERIALS AND METHODS: The ExactVu™ system is currently being used in a multisite, 2,000-patient, randomized clinical trial. Cine loops of 400 biopsies from this trial were used to create the PRI-MUS™ (prostate risk identification using micro-ultrasound) protocol and risk scale. Validation was performed in an independent, pathology blinded set of 100 cines. Three of the 5 investigators performing this validation were familiar with micro-ultrasound but naïve to the PRI-MUS protocol and they received only 1 hour of training. RESULTS: Each increase in risk score demonstrated a 10.1% increase (95% CI 9.3-10.8) in the probability of clinically significant cancer. The risk score also increased with Gleason sum and cancer length with a slope of 0.15 (95% CI 0.09-0.21) and 0.58 (95% CI 0.43-0.73), respectively. Sensitivity and specificity were 80% and 37%, respectively, and the mean ± SD ROC AUC was 60% ± 2%. The protocol was more accurate for detecting high grade disease (Gleason sum greater than 7) with a peak AUC of 74% (mean 66%). CONCLUSIONS: The new resolution of the micro-ultrasound platform paired with the PRI-MUS protocol shows promise for real-time visualization of suspicious lesions and targeting of biopsies. The improved performance of the protocol in more significant disease is consistent with the focus of the field on decreasing insignificant diagnoses and detecting high risk disease early.

Collaborative approach in the development of high-performance brain-computer interfaces for a neuroprosthetic arm: translation from animal models to human control.

Our research group recently demonstrated that a person with tetraplegia could use a brain-computer interface (BCI) to control a sophisticated anthropomorphic robotic arm with skill and speed approaching that of an able-bodied person. This multiyear study exemplifies important principles in translating research from foundational theory and animal experiments into a clinical study. We present a roadmap that may serve as an example for other areas of clinical device research as well as an update on study results. Prior to conducting a multiyear clinical trial, years of animal research preceded BCI testing in an epilepsy monitoring unit, and then in a short-term (28 days) clinical investigation. Scientists and engineers developed the necessary robotic and surgical hardware, software environment, data analysis techniques, and training paradigms. Coordination among researchers, funding institutes, and regulatory bodies ensured that the study would provide valuable scientific information in a safe environment for the study participant. Finally, clinicians from neurosurgery, anesthesiology, physiatry, psychology, and occupational therapy all worked in a multidisciplinary team along with the other researchers to conduct a multiyear BCI clinical study. This teamwork and coordination can be used as a model for others attempting to translate basic science into real-world clinical situations.

Neuroprosthetic technology for individuals with spinal cord injury.

CONTEXT: Spinal cord injury (SCI) results in a loss of function and sensation below the level of the lesion. Neuroprosthetic technology has been developed to help restore motor and autonomic functions as well as to provide sensory feedback. FINDINGS: This paper provides an overview of neuroprosthetic technology that aims to address the priorities for functional restoration as defined by individuals with SCI. We describe neuroprostheses that are in various stages of preclinical development, clinical testing, and commercialization including functional electrical stimulators, epidural and intraspinal microstimulation, bladder neuroprosthesis, and cortical stimulation for restoring sensation. We also discuss neural recording technologies that may provide command or feedback signals for neuroprosthetic devices. CONCLUSION/CLINICAL RELEVANCE: Neuroprostheses have begun to address the priorities of individuals with SCI, although there remains room for improvement. In addition to continued technological improvements, closing the loop between the technology and the user may help provide intuitive device control with high levels of performance.

An electrocorticographic brain interface in an individual with tetraplegia.

Brain-computer interface (BCI) technology aims to help individuals with disability to control assistive devices and reanimate paralyzed limbs. Our study investigated the feasibility of an electrocorticography (ECoG)-based BCI system in an individual with tetraplegia caused by C4 level spinal cord injury. ECoG signals were recorded with a high-density 32-electrode grid over the hand and arm area of the left sensorimotor cortex. The participant was able to voluntarily activate his sensorimotor cortex using attempted movements, with distinct cortical activity patterns for different segments of the upper limb. Using only brain activity, the participant achieved robust control of 3D cursor movement. The ECoG grid was explanted 28 days post-implantation with no adverse effect. This study demonstrates that ECoG signals recorded from the sensorimotor cortex can be used for real-time device control in paralyzed individuals.

Block of peripheral pain response by high-frequency sinusoidal stimulation.

OBJECTIVES: Pain due to peripheral neuropathy is extremely difficult to treat as drugs often become less and less effective over the course of a patient's life. In order to augment such treatments, electrical stimulation has become relatively common, in the form of transcutaneous electrical nerve stimulation, peripheral nerve stimulation, and spinal cord stimulation. Unfortunately, these treatments are only effective in a subset of chronic pain patients. MATERIALS AND METHODS: A rabbit sural-gastrocnemius reflex model of pain was used, in which reflex twitches were elicited using 0.3-Hz 100-µs pulses applied to the sural nerve. This stimulation induces a small electromyogram (EMG) twitch recorded from the gastrocnemius with about 1-msec delay. When pain stimuli are applied on the heel of the foot, the amplitude of the reflex-induced EMG response increased by 4.7 ± 2.5-fold (p < 0.005). Sinusoidal stimulation (high-frequency stimulation) was applied (130 Hz) through a tripolar cuff placed distally on the sural nerve to block the C-fiber activity induced by heel pain. RESULTS: The stimulation paradigm was able to successfully and reversibly block pain signals as measured by the lack of potentiation of the reflex in 100% of the five nerves tested (no significant difference in reflex response, p > 0.5), with thresholds between 500 and 900 µApp . CONCLUSIONS: Complete, reversible block of pain-induced reflex potentiation was obtained in all five nerves tested. This method could be applicable to the control of pain in patients with peripheral neuropathy.

High-performance neuroprosthetic control by an individual with tetraplegia.

BACKGROUND: Paralysis or amputation of an arm results in the loss of the ability to orient the hand and grasp, manipulate, and carry objects, functions that are essential for activities of daily living. Brain-machine interfaces could provide a solution to restoring many of these lost functions. We therefore tested whether an individual with tetraplegia could rapidly achieve neurological control of a high-performance prosthetic limb using this type of an interface. METHODS: We implanted two 96-channel intracortical microelectrodes in the motor cortex of a 52-year-old individual with tetraplegia. Brain-machine-interface training was done for 13 weeks with the goal of controlling an anthropomorphic prosthetic limb with seven degrees of freedom (three-dimensional translation, three-dimensional orientation, one-dimensional grasping). The participant's ability to control the prosthetic limb was assessed with clinical measures of upper limb function. This study is registered with ClinicalTrials.gov, NCT01364480. FINDINGS: The participant was able to move the prosthetic limb freely in the three-dimensional workspace on the second day of training. After 13 weeks, robust seven-dimensional movements were performed routinely. Mean success rate on target-based reaching tasks was 91·6% (SD 4·4) versus median chance level 6·2% (95% CI 2·0-15·3). Improvements were seen in completion time (decreased from a mean of 148 s [SD 60] to 112 s [6]) and path efficiency (increased from 0·30 [0·04] to 0·38 [0·02]). The participant was also able to use the prosthetic limb to do skilful and coordinated reach and grasp movements that resulted in clinically significant gains in tests of upper limb function. No adverse events were reported. INTERPRETATION: With continued development of neuroprosthetic limbs, individuals with long-term paralysis could recover the natural and intuitive command signals for hand placement, orientation, and reaching, allowing them to perform activities of daily living. FUNDING: Defense Advanced Research Projects Agency, National Institutes of Health, Department of Veterans Affairs, and UPMC Rehabilitation Institute.

Hierarchical beamformer and cross-talk reduction in electroneurography.

Electroneurography (ENG) is a method of recording neural activity within nerves. Using nerve electrodes with multiple contacts the activation patterns of individual neuronal fascicles can be estimated by measuring the surface voltages induced by the intraneural activity. The information about neuronal activation can be used for functional electric stimulation (FES) of patients suffering from spinal chord injury, or to control a robotic prosthetic limb of an amputee. However, the ENG signal estimation is a severely ill-posed inverse problem due to uncertainties in the model, low resolution due to limitations of the data, geometric constraints and the difficulty in separating the signal from biological and exogenous noise. In this paper, a reduced computational model for the forward problem is proposed, and the ENG problem is addressed by using beamformer techniques. Furthermore, we show that using a hierarchical statistical model, it is possible to develop an adaptive beamformer algorithm that estimates directly the source variances rather than the voltage source itself. The advantage of this new algorithm, e.g., over a traditional adaptive beamformer algorithm, is that it allows a very stable noise reduction by averaging over a time window. In addition, a new projection technique for separating sources and reducing cross-talk between different fascicle signals is proposed. The algorithms are tested on a computer model of realistic nerve geometry and time series signals.

The impact of electrode characteristics on electrocorticography (ECoG).

Used clinically since Penfield and Jasper's pioneering work in the 1950's, electrocorticography (ECoG) has recently been investigated as a promising technology for brain-computer interfacing. Many researchers have attempted to analyze the properties of ECoG recordings, including prediction of optimal electrode spacing and the improved resolution expected with smaller electrodes. This work applies an analytic model of the volume conductor to investigate the sensitivity field of electrodes of various sizes. The benefit to spatial resolution was minimal for electrodes smaller than ~1mm, while smaller electrodes caused a dramatic decrease in signal-to-noise ratio. The temporal correlation between electrode pairs is predicted over a range of spacings and compared to correlation values from a series of recordings in subjects undergoing monitoring for intractable epilepsy. The observed correlations are found to be much higher than predicted by the analytic model and suggest a more detailed model of cortical activity is needed to identify appropriate ECoG grid spacing.

An algorithm for source signal extraction from the peripheral nerve.

Extracting physiological signals to control external devices such as prosthetics is a field of research that offers great hope for patients suffering from disabilities. In this paper, a novel source signal extraction algorithm, based on the source localization method Champagne, is presented. The algorithm constructs spatial filters that not only maximizes the signal to noise ratio (SNR > 13 dB) of the source activities but also minimizes the cross-talk interference between the sources 10log((P(source of interest)/P(interference sources)) > 14 dB.

Localization and recovery of peripheral neural sources with beamforming algorithms.

The peripheral nervous system carries sensory and motor information that could be useful as command signals for function restoration in areas such as neural prosthetics and functional electrical stimulation (FES). Nerve cuff electrodes provide a robust and safe technique for recording nerve signals. However, a method to separate and recover signals from individual fascicles is necessary. Prior knowledge of the electrode geometry was used to develop an algorithm which assumes neither signal independence nor detailed knowledge of the nerve's geometry/conductivity, and is applicable to any wide-band near-field situation. When used to recover fascicular activities from simulated nerve cuff recordings in a realistic human femoral nerve model, this beamforming algorithm separates signals as close as 1.5 mm with cross-correlation coefficient, R > 0.9 (10% noise). Ten simultaneous signals could be recovered from individual fascicles with only a 20% decrease in R compared to a single signal. At high noise levels (40%), sources were localized to 180 +/- 170 microm in the 12 x 3 mm cuff. Localizing sources and using the resulting positions in the recovery algorithm yielded R = 0.66 +/- 0.10 in 10% noise for five simultaneous muscle-activation signals from synergistic fascicles. These recovered signals should allow natural, robust, closed-loop control of multiple degree-of-freedom prosthetic devices and FES systems.