Precise Ground Truth Research Articles

Baby steps in the path of modifying the role of cardiologists for interpreting EKG for AMI

Abstract Background EKG interpretation is slowly transitioning to a physician-free, Artificial Intelligence (AI)-driven endeavor. Our continued efforts to innovate follow a carefully laid stepwise approach, as follows: 1) Create an AI algorithm that accurately identifies STEMI against non-STEMI using a 12-lead EKG; 2) Challenging said algorithm by including different EKG diagnosis to the previous experiment, and now 3) To further validate the accuracy and reliability of our algorithm while also improving performance in a prehospital and hospital settings. Purpose To provide an accurate, reliable, and cost-effective tool for STEMI detection with the potential to redirect human resources into other clinically relevant tasks and save the need for human resources. Methods Database: EKG records obtained from Latin America Telemedicine Infarct Network (Mexico, Colombia, Argentina, and Brazil) from April 2014 to December 2019. Dataset: A total of 11,567 12-lead EKG records of 10-seconds length with sampling frequency of 500 [Hz], including the following balanced classes: unconfirmed and angiographically confirmed STEMI, branch blocks, non-specific ST-T abnormalities, normal and abnormal (200+ CPT codes, excluding the ones included in other classes). The label of each record was manually checked by cardiologists to ensure precision (Ground truth). Pre-processing: The first and last 250 samples were discarded as they may contain a standardization pulse. An order 5 digital low pass filter with a 35 Hz cut-off was applied. For each record, the mean was subtracted to each individual lead. Classification: The determined classes were STEMI (STEMI in different locations of the myocardium – anterior, inferior and lateral); Not-STEMI (A combination of randomly sampled normal, branch blocks, non-specific ST-T abnormalities and abnormal records – 25% of each subclass). Training & Testing: A 1-D Convolutional Neural Network was trained and tested with a dataset proportion of 90/10; respectively. The last dense layer outputs a probability for each record of being STEMI or Not-STEMI. Additional testing was performed with a subset of the original dataset of angiographically confirmed STEMI. Results See Figure Attached – Preliminary STEMI Dataset Accuracy: 96.4%; Sensitivity: 95.3%; Specificity: 97.4% – Confirmed STEMI Dataset: Accuracy: 97.6%; Sensitivity: 98.1%; Specificity: 97.2%. Conclusions Our results remain consistent with our previous experience. By further increasing the amount and complexity of the data, the performance of the model improves. Future implementations of this technology in clinical settings look promising, not only in performing swift screening and diagnostic steps but also partaking in complex STEMI management triage. Funding Acknowledgement Type of funding source: None

Innovative techniques to construct powerful artificial intelligence algorithms for st-elevation myocardial infarction

Abstract Background With the sudden advent of Artificial Intelligence (AI), incorporation of these technologies into key aspects of our working environment has become an ever so delicate task, especially so when dealing with time-sensitive and potentially lethal scenarios such as ST-Elevation Myocardial Infarction (STEMI) management. By further expanding into our successful experiences with AI-guided algorithms for STEMI detection, we implemented an innovative ensemble method into our methodology as we seek to improve the algorithm's predictive capabilities. Purpose Through the ensemble method, we combined two ML techniques to boost our previous experiments' accuracy and reliability. Methods Database: EKG records obtained from Latin America Telemedicine Infarct Network (Mexico, Colombia, Argentina, and Brazil) from April 2014 to December 2019. Dataset: Two separate datasets were used to train and test two sets of AI algorithms. The first comprised of 11,567 records and the second 7,286 records, each composed of 12-lead EKG records of 10-second length with sampling frequency of 500 Hz, including the following balanced classes: unconfirmed & angiographically confirmed STEMI (first model); angiographically confirmed STEMI only (second model); and, for both models, we included branch blocks, non-specific ST-T abnormalities, normal, and abnormal (200+ CPT codes, excluding the ones included in other classes). Label per record was manually checked by cardiologists to ensure precision (Ground truth). Pre-processing: First and last 250 samples were discarded to avoid a standardization pulse. An order 5 digital low pass filter with a 35 Hz cut-off was applied. For each record, the mean was subtracted to each individual lead. Classification: The determined classes were STEMI and Not-STEMI (A combination of randomly sampled normal, branch blocks, non-specific ST-T abnormalities and abnormal records – 25% of each subclass). Training & Testing: The last dense layer outputs a probability for each record of being STEMI or Not-STEMI. These probabilities were calculated for each model (Model 1 trained with Complete STEMI dataset and Model 2 trained with confirmed STEMI only dataset) and aggregated using the mean aggregation to generate the final label for each record. A 1-D Convolutional Neural Network was trained and tested with a dataset proportion of 90%/10%; respectively. Results are reported for both testing datasets (Complete and confirmed STEMI only records). Results Complete STEMI Dataset: Accuracy: 96.5% Sensitivity: 96.2% Specificity: 96.9% – Confirmed STEMI only Dataset: Accuracy: 98.5% Sensitivity: 98.3% Specificity: 98.6%' Conclusion(s) While Model 1 and Model 2 achieved similar performances with promising results on their own, applying a combination of both through the ensemble model exhibits a clear improvement in performance when applied to both datasets. This provides a blueprint for advanced automated STEMI detection through wearable devices. Funding Acknowledgement Type of funding source: None

Enhancing AI-guided STEMI detection algorithms by incorporating higher quality fiduciary EKG elements

Abstract Background As EKG interpretation paradigms to a physician-free milieu, accumulating massive quantities of distilled pre-processed data becomes a must for machine learning techniques. In our pursuit of reducing ischemic times in STEMI management, we have improved our Artificial Intelligence (AI)-guided diagnostic tool by following a three-step approach: 1) Increase accuracy by adding larger clusters of data. 2) Increase the breadth of EKG classifications to provide more precise feedback and further refine the inputs which ultimately reflects in better and more accurate outputs. 3) Improving the algorithms' ability to discern between cardiovascular entities reflected in the EKG records. Purpose To bolster our algorithm's accuracy and reliability for electrocardiographic STEMI recognition. Methods Dataset: A total of 7,286 12-lead EKG records of 10-seconds length with a sampling frequency of 500 Hz obtained from Latin America Telemedicine Infarct Network from April 2014 to December 2019. This included the following balanced classes: angiographically confirmed STEMI, branch blocks, non-specific ST-T abnormalities, normal, and abnormal (200+ CPT codes, excluding the ones included in other classes). Labels of each record were manually checked by cardiologists to ensure precision (Ground truth). Pre-processing: First and last 250 samples were discarded to avoid a standardization pulse. Order 5 digital low pass filters with a 35 Hz cut-off was applied. For each record, the mean was subtracted to each individual lead. Classification: Determined classes were “STEMI” and “Not-STEMI” (A combination of randomly sampled normal, branch blocks, non-specific ST-T abnormalities and abnormal records – 25% of each subclass). Training & Testing: A 1-D Convolutional Neural Network was trained and tested with a dataset proportion of 90/10, respectively. The last dense layer outputs a probability for each record of being STEMI or Not-STEMI. Additional testing was performed with a subset of the original complete dataset of unconfirmed STEMI. Performance indicators (accuracy, sensitivity, and specificity) were calculated for each model and results were compared with our previous findings from past experiments. Results Complete STEMI data: Accuracy: 95.9% Sensitivity: 95.7% Specificity: 96.5%; Confirmed STEMI: Accuracy: 98.1% Sensitivity: 98.1% Specificity: 98.1%; Prior Data obtained in our previous experiments are shown below for comparison. Conclusion(s) After the addition of clustered pre-processed data, all performance indicators for STEMI detection increased considerably between both Confirmed STEMI datasets. On the other hand, the Complete STEMI dataset kept a strong and steady set of performance metrics when compared with past results. These findings not only validate the consistency and reliability of our algorithm but also connotes the importance of creating a pristine dataset for this and any other AI-derived medical tools. Funding Acknowledgement Type of funding source: None

A LOW-DELAY AND ROBUST SOLUTION FOR RAPID MOVEMENT DETECTION USING UWB

Abstract. The increasing applications for location-based services (LBS) and real-time location system (RTLS) promote the development of indoor positioning and navigation technology. However, the positioning methods based on traditional Radio Frequency (RF) signals are difficult to improve the accuracy and sensitivity for rapid movement detection with a low latency parameter. In order to solve those problems, this paper adopted ultra-wideband (UWB) chips as the wireless communication technology and designed a method of rapid movement detection with low delay. The method consists of a topology of combined two way ranging (C-TWR), a schedule of time division multiple access (TDMA), an non-light-of-sight (NLOS) handling mechanism and an adaptive extended Kalman filter (A-EKF). To examine the low delay performance, accuracy and robustness of the system, a specific electro-displacement actuator with programmable speed is applied to provide precise ground truth for real speed and distance, and a system test platform consisting of four anchors and four tags provides information of specific situation. The results show that the system can locate the rapid moving pedestrian with a maximal speed of approximately 5 m s−1 in real-time with a positioning accuracy of less than 0.5 m. There are profound implications for further studies in the range of human wearable tags.

DFR: Differentiable Function Rendering for Learning 3D Generation from Images

AbstractLearning‐based 3D generation is a popular research field in computer graphics. Recently, some works adapted implicit function defined by a neural network to represent 3D objects and have become the current state‐of‐the‐art. However, training the network requires precise ground truth 3D data and heavy pre‐processing, which is unrealistic. To tackle this problem, we propose the DFR, a differentiable process for rendering implicit function representation of 3D objects into 2D images. Briefly, our method is to simulate the physical imaging process by casting multiple rays through the image plane to the function space, aggregating all information along with each ray, and performing a differentiable shading according to every ray's state. Some strategies are also proposed to optimize the rendering pipeline, making it efficient both in time and memory to support training a network. With DFR, we can perform many 3D modeling tasks with only 2D supervision. We conduct several experiments for various applications. The quantitative and qualitative evaluations both demonstrate the effectiveness of our method.

Score Following as a Multi-Modal Reinforcement Learning Problem

Score following is the process of tracking a musical performance (audio) in a corresponding symbolic representation (score). While methods using computer-readable score representations as input are able to achieve reliable tracking results, there is little research on score following based on raw score images. In this paper, we build on previous work that formulates the score following task as a multi-modal Markov Decision Process (MDP). Given this formal definition, one can address the problem of score following with state-of-the-art deep reinforcement learning (RL) algorithms. In particular, we design end-to-end multi-modal RL agents that simultaneously learn to listen to music recordings, read the scores from images of sheet music, and follow the music along in the sheet. Using algorithms such as synchronous Advantage Actor Critic (A2C) and Proximal Policy Optimization (PPO), we reproduce and further improve existing results. We also present first experiments indicating that this approach can be extended to track real piano recordings of human performances. These audio recordings are made openly available to the research community, along with precise note-level alignment ground truth.

Epithelium segmentation using deep learning in H&E-stained prostate specimens with immunohistochemistry as reference standard

Given the importance of gland morphology in grading prostate cancer (PCa), automatically differentiating between epithelium and other tissues is an important prerequisite for the development of automated methods for detecting PCa. We propose a new deep learning method to segment epithelial tissue in digitised hematoxylin and eosin (H&E) stained prostatectomy slides using immunohistochemistry (IHC) as reference standard. We used IHC to create a precise and objective ground truth compared to manual outlining on H&E slides, especially in areas with high-grade PCa. 102 tissue sections were stained with H&E and subsequently restained with P63 and CK8/18 IHC markers to highlight epithelial structures. Afterwards each pair was co-registered. First, we trained a U-Net to segment epithelial structures in IHC using a subset of the IHC slides that were preprocessed with color deconvolution. Second, this network was applied to the remaining slides to create the reference standard used to train a second U-Net on H&E. Our system accurately segmented both intact glands and individual tumour epithelial cells. The generalisation capacity of our system is shown using an independent external dataset from a different centre. We envision this segmentation as the first part of a fully automated prostate cancer grading pipeline.

Evaluating the Performance of Structure from Motion Pipelines

Structure from Motion (SfM) is a pipeline that allows three-dimensional reconstruction starting from a collection of images. A typical SfM pipeline comprises different processing steps each of which tackles a different problem in the reconstruction pipeline. Each step can exploit different algorithms to solve the problem at hand and thus many different SfM pipelines can be built. How to choose the SfM pipeline best suited for a given task is an important question. In this paper we report a comparison of different state-of-the-art SfM pipelines in terms of their ability to reconstruct different scenes. We also propose an evaluation procedure that stresses the SfM pipelines using real dataset acquired with high-end devices as well as realistic synthetic dataset. To this end, we created a plug-in module for the Blender software to support the creation of synthetic datasets and the evaluation of the SfM pipeline. The use of synthetic data allows us to easily have arbitrarily large and diverse datasets with, in theory, infinitely precise ground truth. Our evaluation procedure considers both the reconstruction errors as well as the estimation errors of the camera poses used in the reconstruction.

Attitude estimation for indoor navigation and augmented reality with smartphones

We investigate the precision of attitude estimation algorithms in the particular context of pedestrian navigation with commodity smartphones and their inertial/magnetic sensors. We report on an extensive comparison and experimental analysis of existing algorithms. We focus on typical motions of smartphones when carried by pedestrians. We use a precise ground truth obtained from a motion capture system. We test state-of-the-art and built-in attitude estimation techniques with several smartphones, in the presence of magnetic perturbations typically found in buildings. We discuss the obtained results, analyze advantages and limits of current technologies for attitude estimation in this context. Furthermore, we propose a new technique for limiting the impact of magnetic perturbations with any attitude estimation algorithm used in this context. We show how our technique compares and improves over previous works. A particular attention was paid to the study of attitude estimation in the context of augmented reality motions when using smartphones.

Android Malware Detection with Weak Ground Truth Data

For Android malware detection, precise ground truth is a rare commodity. As security knowledge evolves, what may be considered ground truth at one moment in time may change, and apps once considered benign may turn out to be malicious. The inevitable noise in data labels poses a challenge to inferring effective machine learning classifiers. Our work is focused on approaches for learning classifiers for Android malware detection in a manner that is methodologically sound with regard to the uncertain and ever-changing ground truth in the problem space. We leverage the fact that although data labels are unavoidably noisy, a malware label is much more precise than a benign label. While you can be confident that an app is malicious, you can never be certain that a benign app is really benign, or just undetected malware. Based on this insight, we leverage a modified Logistic Regression classifier that allows us to learn from only positive and unlabeled data, without making any assumptions about benign labels. We find Label Regularized Logistic Regression to perform well for noisy app datasets, as well as datasets where there is a limited amount of positive labeled data, both of which are representative of real-world situations.

ON CONSTRUCTION OF A RELIABLE GROUND TRUTH FOR EVALUATION OF VISUAL SLAM ALGORITHMS

In this work we are concerning the problem of localization accuracy evaluation of visual-based Simultaneous Localization and Mapping (SLAM) techniques. Quantitative evaluation of the SLAM algorithm performance is usually done using the established metrics of Relative pose error and Absolute trajectory error which require a precise and reliable ground truth. Such a ground truth is usually hard to obtain, while it requires an expensive external localization system. In this work we are proposing to use the SLAM algorithm itself to construct a reliable ground-truth by offline frame-by-frame processing. The generated ground-truth is suitable for evaluation of different SLAM systems, as well as for tuning the parametrization of the on-line SLAM. The presented practical experimental results indicate the feasibility of the proposed approach.

Online Body Schema Adaptation Based on Internal Mental Simulation and Multisensory Feedback

In this paper, we describe a novel approach to obtain automatic adaptation of the robot body schema and to improve the robot perceptual and motor skills based on this body knowledge. Predictions obtained through a mental simulation of the body are combined with the real sensory feedback to achieve two objectives simultaneously: body schema adaptation and markerless 6D hand pose estimation. The body schema consists of a computer graphics simulation of the robot, which includes the arm and head kinematics (adapted online during the movements) and an appearance model of the hand shape and texture. The mental simulation process generates predictions on how the hand will appear in the robot camera images, based on the body schema and the proprioceptive information (i.e. motor encoders). These predictions are compared to the actual images using Sequential Monte Carlo techniques to feed a particle-based Bayesian estimation method to estimate the parameters of the body schema. The updated body schema will improve the estimates of the 6D hand pose, which is then used in a closed-loop control scheme (i.e. visual servoing), enabling precise reaching. We report experiments with the iCub humanoid robot that support the validity of our approach. A number of simulations with precise ground-truth were performed to evaluate the estimation capabilities of the proposed framework. Then, we show how the use of high-performance GPU programming and an edge-based algorithm for visual perception allow for real-time implementation in real world scenarios.

Exploiting View-Specific Appearance Similarities Across Classes for Zero-Shot Pose Prediction: A Metric Learning Approach

Viewpoint estimation, especially in case of multiple object classes, remains an important and challenging problem. First, objects under different views undergo extreme appearance variations, often making within-class variance larger than between-class variance. Second, obtaining precise ground truth for real-world images, necessary for training supervised viewpoint estimation models, is extremely difficult and time consuming. As a result, annotated data is often available only for a limited number of classes. Hence it is desirable to share viewpoint information across classes. Additional complexity arises from unaligned pose labels between classes, i.e. a side view of a car might look more like a frontal view of a toaster, than its side view. To address these problems, we propose a metric learning approach for joint class prediction and pose estimation. Our approach allows to circumvent the problem of viewpoint alignment across multiple classes, and does not require dense viewpoint labels. Moreover, we show, that the learned metric generalizes to new classes, for which the pose labels are not available, and therefore makes it possible to use only partially annotated training sets, relying on the intrinsic similarities in the viewpoint manifolds. We evaluate our approach on two challenging multi-class datasets, 3DObjects and PASCAL3D+.

Non-parametric calibration for depth sensors

RGBD sensors are commonly used in robotics applications for many purposes, including 3D reconstruction of the environment and mapping. In these tasks, uncalibrated sensors can generate poor quality results. In this paper we propose a quick and easy to use approach to estimate the undistortion function of RGBD sensors. Our approach does not rely on the knowledge of the sensor model, on the use of a specific calibration pattern or on external SLAM systems to track the device position. We compute an extensive representation of the undistortion function as well as its statistics and use machine learning methods for approximation of the undistortion function. We validated our approach on datasets acquired from different kinds of RGBD sensors and using a precise 3D ground truth. We also provide a procedure for evaluating the quality of the calibration using a mobile robot and a 2D laser range finder. The results clearly show the advantages in using sensor data calibrated with the method described in this paper.

IMPROVEMENT OF 3D MONTE CARLO LOCALIZATION USING A DEPTH CAMERA AND TERRESTRIAL LASER SCANNER

Abstract. Effective and accurate localization method in three-dimensional indoor environments is a key requirement for indoor navigation and lifelong robotic assistance. So far, Monte Carlo Localization (MCL) has given one of the promising solutions for the indoor localization methods. Previous work of MCL has been mostly limited to 2D motion estimation in a planar map, and a few 3D MCL approaches have been recently proposed. However, their localization accuracy and efficiency still remain at an unsatisfactory level (a few hundreds millimetre error at up to a few FPS) or is not fully verified with the precise ground truth. Therefore, the purpose of this study is to improve an accuracy and efficiency of 6DOF motion estimation in 3D MCL for indoor localization. Firstly, a terrestrial laser scanner is used for creating a precise 3D mesh model as an environment map, and a professional-level depth camera is installed as an outer sensor. GPU scene simulation is also introduced to upgrade the speed of prediction phase in MCL. Moreover, for further improvement, GPGPU programming is implemented to realize further speed up of the likelihood estimation phase, and anisotropic particle propagation is introduced into MCL based on the observations from an inertia sensor. Improvements in the localization accuracy and efficiency are verified by the comparison with a previous MCL method. As a result, it was confirmed that GPGPU-based algorithm was effective in increasing the computational efficiency to 10-50 FPS when the number of particles remain below a few hundreds. On the other hand, inertia sensor-based algorithm reduced the localization error to a median of 47mm even with less number of particles. The results showed that our proposed 3D MCL method outperforms the previous one in accuracy and efficiency.

SAR Imagery Simulation of Ship Based on Electromagnetic Calculations and Sea Clutter Modelling for Classification Applications

Ship detection and classification with space-borne SAR has many potential applications within the maritime surveillance, fishery activity management, monitoring ship traffic, and military security. While ship detection techniques with SAR imagery are well established, ship classification is still an open issue. One of the main reasons may be ascribed to the difficulties on acquiring the required quantities of real data of vessels under different observation and environmental conditions with precise ground truth. Therefore, simulation of SAR images with high scenario flexibility and reasonable computation costs is compulsory for ship classification algorithms development. However, the simulation of SAR imagery of ship over sea surface is challenging. Though great efforts have been devoted to tackle this difficult problem, it is far from being conquered. This paper proposes a novel scheme for SAR imagery simulation of ship over sea surface. The simulation is implemented based on high frequency electromagnetic calculations methods of PO, MEC, PTD and GO. SAR imagery of sea clutter is modelled by the representative K-distribution clutter model. Then, the simulated SAR imagery of ship can be produced by inserting the simulated SAR imagery chips of ship into the SAR imagery of sea clutter. The proposed scheme has been validated with canonical and complex ship targets over a typical sea scene.

Comparative evaluation of range sensor accuracy for indoor mobile robotics and automated logistics applications

3D range sensing is an important topic in robotics, as it is a component in vital autonomous subsystems such as for collision avoidance, mapping and perception. The development of affordable, high frame rate and precise 3D range sensors is thus of considerable interest. Recent advances in sensing technology have produced several novel sensors that attempt to meet these requirements. This work is concerned with the development of a holistic method for accuracy evaluation of the measurements produced by such devices. A method for comparison of range sensor output to a set of reference distance measurements, without using a precise ground truth environment model, is proposed. This article presents an extensive evaluation of three novel depth sensors — the Swiss Ranger SR-4000, Fotonic B70 and Microsoft Kinect. Tests are concentrated on the automated logistics scenario of container unloading. Six different setups of box-, cylinder-, and sack-shaped goods inside a mock-up container are used to collect range measurements. Comparisons are performed against hand-crafted ground truth data, as well as against a reference actuated Laser Range Finder (aLRF) system. Additional test cases in an uncontrolled indoor environment are performed in order to evaluate the sensors’ performance in a challenging, realistic application scenario.

Bias Estimation and Calibration of a Pair of Laser Range Sensors

Abstract Laser range sensors (LRS) are ubiquitous in mobile robotics. They are usually modeled as having statistical error only, despite some having systematic error (bias) documented. This paper deals with bias estimation for laser range sensors in order to produce better calibration for two LRS. Assuming both LRS sense in the same plane, we propose an algorithm to estimate position and orientation of the second laser in the coordinate frame of the first laser, or vice versa. In order to truly have corresponding points in two LRS scans, we propose a calibration target. We use cylindrical objects of known radius, placed in the field of view of both LRS, perpendicularly to the sensing plane. Centers of circles that result from the intersection of the sensing plane and cylindrical objects then become corresponding points. Our algorithm first estimates cylinder centers, together with their respective uncertainties, and then using those results produces Euclidean transform estimate, together with it's respective uncertainty. The results of this algorithm can be used to gather very precise ground truth for people tracking applications with LRS. In case of occlusions, one has obvious benefits from two or more different viewpoints.

Benchmarking template-based tracking algorithms

For natural interaction with augmented reality (AR) applications, good tracking technology is key. But unlike dense stereo, optical flow or multi-view stereo, template-based tracking which is most commonly used for AR applications lacks benchmark datasets allowing a fair comparison between state-of-the-art algorithms. Until now, in order to evaluate objectively and quantitatively the performance and the robustness of template-based tracking algorithms, mainly synthetically generated image sequences were used. The evaluation is therefore often intrinsically biased. In this paper, we describe the process we carried out to perform the acquisition of real-scene image sequences with very precise and accurate ground truth poses using an industrial camera rigidly mounted on the end effector of a high-precision robotic measurement arm. For the acquisition, we considered most of the critical parameters that influence the tracking results such as: the texture richness and the texture repeatability of the objects to be tracked, the camera motion and speed, and the changes of the object scale in the images and variations of the lighting conditions over time. We designed an evaluation scheme for object detection and interframe tracking algorithms suited for AR and other computer vision applications and used the image sequences to apply this scheme to several state-of-the-art algorithms. The image sequences are freely available for testing, submitting and evaluating new template-based tracking algorithms, i.e. algorithms that detect or track a planar object in an image sequence given only one image of the object (called the template).

A hierarchical spectral clustering and nonlinear dimensionality reduction scheme for detection of prostate cancer from magnetic resonance spectroscopy (MRS)

Magnetic resonance spectroscopy (MRS) has been shown to have great clinical potential as a supplement to magnetic resonance imaging in the detection of prostate cancer (CaP). MRS provides functional information in the form of changes in the relative concentration of specific metabolites including choline, creatine, and citrate which can be used to identify potential areas of CaP. With a view to assisting radiologists in interpretation and analysis of MRS data, some researchers have begun to develop computer-aided detection (CAD) schemes for CaP identification from spectroscopy. Most of these schemes have been centered on identifying and integrating the area under metabolite peaks which is then used to compute relative metabolite ratios. However, manual identification of metabolite peaks on the MR spectra, and especially via CAD, is a challenging problem due to low signal-to-noise ratio, baseline irregularity, peak overlap, and peak distortion. In this article the authors present a novel CAD scheme that integrates nonlinear dimensionality reduction (NLDR) with an unsupervised hierarchical clustering algorithm to automatically identify suspicious regions on the prostate using MRS and hence avoids the need to explicitly identify metabolite peaks. The methodology comprises two stages. In stage 1, a hierarchical spectral clustering algorithm is used to distinguish between extracapsular and prostatic spectra in order to localize the region of interest (ROI) corresponding to the prostate. Once the prostate ROI is localized, in stage 2, a NLDR scheme, in conjunction with a replicated clustering algorithm, is used to automatically discriminate between three classes of spectra (normal appearing, suspicious appearing, and indeterminate). The methodology was quantitatively and qualitatively evaluated on a total of 18 1.5 T in vivo prostate T2-weighted (w) and MRS studies obtained from the multisite, multi-institutional American College of Radiology (ACRIN) trial. In the absence of the precise ground truth for CaP extent on the MR imaging for most of the ACRIN studies, probabilistic quantitative metrics were defined based on partial knowledge on the quadrant location and size of the tumor. The scheme, when evaluated against this partial ground truth, was found to have a CaP detection sensitivity of 89.33% and specificity of 79.79%. The results obtained from randomized threefold and fivefold cross validation suggest that the NLDR based clustering scheme has a higher CaP detection accuracy compared to such commonly used MRS analysis schemes as z score and PCA. In addition, the scheme was found to be robust to changes in system parameters. For 6 of the 18 studies an expert radiologist laboriously labeled each of the individual spectra according to a five point scale, with 1/2 representing spectra that the expert considered normal and 3/4/5 being spectra the expert deemed suspicious. When evaluated on these expert annotated datasets, the CAD system yielded an average sensitivity (cluster corresponding to suspicious spectra being identified as the CaP class) and specificity of 81.39% and 64.71%, respectively.