Improvement In Overall Accuracy Research Articles

Impervious surface extraction in imbalanced datasets: integrating partial results and multi-temporal information in an iterative one-class classifier

ABSTRACTAccurate urban land use/cover monitoring is an essential step towards a sustainable future. As a key part of the classification process, the characteristics of reference data can significantly affect classification accuracy and quality of produced maps. However, ideal reference data is not always readily available; users frequently have difficulty generating sufficient reference data for some classes given time, cost, data availability, expertise level, or other limitations. This study aims at dealing with this lack of sufficiently balanced reference data by presenting a modified hybrid one-class support vector data description (SVDD) model. The underlying hypothesis is that the lack of balanced reference data can be overcome through integration of partially extracted results and multi-temporal spectral information. The partially extracted results, defined as highly accurate classified pixels identified in previous algorithmic iterations, allow a gradual increase of the available training data. Furthermore, the method incorporates a voting system that integrates multi-temporal images using the SVDD algorithm. We applied this hybrid method to binary impervious classification of multi-temporal Landsat Thematic Mapper imagery from Central New York with imbalanced reference data. The proposed hybrid one-class SVDD model achieved a 5–6% improvement in overall accuracy and 0.05–0.09 in kappa than the typical one-class SVDD benchmark. While the method was tested on a single site (albeit with an unusually high reference dataset size of >870,000 pixels) we feel confident to suggest implementation of our methodology in other sites over the traditional method. This is because our approach automatically reverts to the traditional method when voting is inconsistent or there is a limited number of highly accurately classified pixels to assist future iterations. Future work could explore the quantity and temporal specificity (e.g. benefits of specific months) of the multi-temporal image selection and/or test other one-class classifiers.

Granular computing-based approach for classification towards reduction of bias in ensemble learning

Machine learning has become a powerful approach in practical applications, such as decision making, sentiment analysis and ontology engineering. To improve the overall performance in machine learning tasks, ensemble learning has become increasingly popular by combining different learning algorithms or models. Popular approaches of ensemble learning include Bagging and Boosting, which involve voting towards the final classification. The voting in both Bagging and Boosting could result in incorrect classification due to the bias in the way voting takes place. To reduce the bias in voting, this paper proposes a probabilistic approach of voting in the context of granular computing towards improvement of overall accuracy of classification. An experimental study is reported to validate the proposed approach of voting using 15 data sets from the UCI repository. The results show that probabilistic voting is effective in increasing the accuracy through reduction of the bias in voting. This paper contributes to the theoretical and empirical analysis of causes of bias in voting, towards advancing ensemble learning approaches through the use of probabilistic voting.

Pediatric Disaster Triage: Multiple Simulation Curriculum Improves Prehospital Care Providers' Assessment Skills

ABSTRACTObjective: Paramedics and emergency medical technicians (EMTs) triage pediatric disaster victims infrequently. The objective of this study was to measure the effect of a multiple-patient, multiple-simulation curriculum on accuracy of pediatric disaster triage (PDT). Methods: Paramedics, paramedic students, and EMTs from three sites were enrolled. Triage accuracy was measured three times (Time 0, Time 1 [two weeks later], and Time 2 [6 months later]) during a disaster simulation, in which high and low fidelity manikins and actors portrayed 10 victims. Accuracy was determined by participant triage decision concordance with predetermined expected triage level (RED [Immediate], YELLOW [Delayed], GREEN [Ambulatory], BLACK [Deceased]) for each victim. Between Time 0 and Time 1, participants completed an interactive online module, and after each simulation there was an individual debriefing. Associations between participant level of training, years of experience, and enrollment site were determined, as were instances of the most dangerous mistriage, when RED and YELLOW victims were triaged BLACK. Results: The study enrolled 331 participants, and the analysis included 261 (78.9%) participants who completed the study, 123 from the Connecticut site, 83 from Rhode Island, and 55 from Massachusetts. Triage accuracy improved significantly from Time 0 to Time 1, after the educational interventions (first simulation with debriefing, and an interactive online module), with a median 10% overall improvement (p < 0.001). Subgroup analyses showed between Time 0 and Time 1, paramedics and paramedic students improved more than EMTs (p = 0.002). Analysis of triage accuracy showed greatest improvement in overall accuracy for YELLOW triage patients (Time 0 50% accurate, Time1 100%), followed by RED patients (Time 0 80%, Time 1 100%). There was no significant difference in accuracy between Time 1 and Time 2 (p = 0.073). Conclusion: This study shows that the multiple-victim, multiple-simulation curriculum yields a durable 10% improvement in simulated triage accuracy. Future iterations of the curriculum can target greater improvements in EMT triage accuracy.

Assessment of Compliance With RF EMF Exposure Limits: Approximate Methods for Radio Base Station Products Utilizing Array Antennas With Beam-Forming Capabilities

The importance of antenna arrays with beam-forming capability is expected to increase in future mobile communication systems with the ongoing development of new radio access technologies involving higher frequency bands and massive MIMO concepts. In this paper, various approximate methods for radio frequency electromagnetic field compliance assessments of radio base station products utilizing these antenna arrays are investigated with the focus on front compliance distances. For electrically large arrays, the results show that accurate compliance boundary dimensions may be obtained using a field-combining procedure of assessment results for a centrally placed element. For smaller arrays, an overall improvement in accuracy is possible by complementing the center element solution with results for a handful of selected edge and corner elements to better characterize the array behavior. Compared with a general and straightforward approach based on field combining of embedded assessment results for each element, the approximate methods allow for significant reductions of the total assessment time.

Flux renormalization in constant power burnup calculations

To more accurately represent the desired power in a constant power burnup calculation, the depletion steps of the calculation can be divided into substeps and the neutron flux renormalized on each substep to match the desired power. This paper explores how such renormalization should be performed, how large a difference it makes, and whether using renormalization affects results regarding the relative performance of different neutronics–depletion coupling schemes. When used with older coupling schemes, renormalization can provide a considerable improvement in overall accuracy. With previously published higher order coupling schemes, which are more accurate to begin with, renormalization has a much smaller effect. While renormalization narrows the differences in the accuracies of different coupling schemes, their order of accuracy is not affected.

A COMPARISON OF SUB-PIXEL MAPPING METHODS FOR COASTAL AREAS

Abstract. This paper presents the comparisons of three soft classification methods and three sub-pixel mapping methods for the classification of coastal areas at sub-pixel level. Specifically, SPOT-7 multispectral images covering the coastal area of Perth are selected as the experiment dataset. For the soft classification, linear spectral unmixing model, supervised fully-fuzzy classification method and the support vector machine are applied to generate the fraction map. Then for the sub-pixel mapping, the sub-pixel/pixel attraction model, pixel swapping and wavelets method are compared. Besides, the influence of the correct fraction constraint is explored. Moreover, a post-processing step is implemented according to the known spatial knowledge of coastal areas. The accuracy assessment of the fraction values indicates that support vector machine generates the most accurate fraction result. For sub-pixel mapping, wavelets method outperforms the other two methods with overall classification accuracy of 91.79% and Kappa coefficient of 0.875 after the post-processing step and it also performs best for waterline extraction with mean distance of 0.71m to the reference waterline. In this experiment, the use of correct fraction constraint decreases the classification accuracy of sub-pixel mapping methods and waterline extraction. Finally, the post-processing step improves the accuracy of sub-pixel mapping methods, especially for those with correct coefficient constraint. The most significant improvement of overall accuracy is as much as 4% for the sub-pixel/pixel attraction model with correct coefficient constraint.

A COMPARISON OF SUB-PIXEL MAPPING METHODS FOR COASTAL AREAS

This paper presents the comparisons of three soft classification methods and three sub-pixel mapping methods for the classification of coastal areas at sub-pixel level. Specifically, SPOT-7 multispectral images covering the coastal area of Perth are selected as the experiment dataset. For the soft classification, linear spectral unmixing model, supervised fully-fuzzy classification method and the support vector machine are applied to generate the fraction map. Then for the sub-pixel mapping, the sub-pixel/pixel attraction model, pixel swapping and wavelets method are compared. Besides, the influence of the correct fraction constraint is explored. Moreover, a post-processing step is implemented according to the known spatial knowledge of coastal areas. The accuracy assessment of the fraction values indicates that support vector machine generates the most accurate fraction result. For sub-pixel mapping, wavelets method outperforms the other two methods with overall classification accuracy of 91.79% and Kappa coefficient of 0.875 after the post-processing step and it also performs best for waterline extraction with mean distance of 0.71m to the reference waterline. In this experiment, the use of correct fraction constraint decreases the classification accuracy of sub-pixel mapping methods and waterline extraction. Finally, the post-processing step improves the accuracy of sub-pixel mapping methods, especially for those with correct coefficient constraint. The most significant improvement of overall accuracy is as much as 4% for the sub-pixel/pixel attraction model with correct coefficient constraint.

A meta-analysis of remote sensing research on supervised pixel-based land-cover image classification processes: General guidelines for practitioners and future research

Abstract Classification of remotely sensed imagery for land-cover mapping purposes has attracted significant attention from researchers and practitioners. Numerous studies conducted over several decades have investigated a broad array of input data and classification methods. However, this vast assemblage of research results has not been synthesized to provide coherent guidance on the relative performance of different classification processes for generating land cover products. To address this problem, we completed a statistical meta-analysis of the past 15 years of research on supervised per-pixel image classification published in five high-impact remote sensing journals. The two general factors evaluated were classification algorithms and input data manipulation as these are factors that can be controlled by analysts to improve classification accuracy. The meta-analysis revealed that inclusion of texture information yielded the greatest improvement in overall accuracy of land-cover classification with an average increase of 12.1%. This increase in accuracy can be attributed to the additional spatial context information provided by including texture. Inclusion of ancillary data, multi-angle and time images also provided significant improvement in classification overall accuracy, with 8.5%, 8.0%, and 6.9% of average improvements, respectively. In contrast, other manipulation of spectral information such as index creation (e.g. Normalized Difference Vegetation Index) and feature extraction (e.g. Principal Components Analysis) offered much smaller improvements in accuracy. In terms of classification algorithms, support vector machines achieved the greatest accuracy, followed by neural network methods. The random forest classifier performed considerably better than the traditional decision tree classifier. Maximum likelihood classifiers, often used as benchmarking algorithms, offered low accuracy. Our findings will help guide practitioners to decide which classification to implement and also provide direction to researchers regarding comparative studies that will further solidify our understanding of different classification processes. However, these general guidelines do not preclude an analyst from incorporating personal preferences or considering specific algorithmic benefits that may be pertinent to a particular application.

Discrimination of tropical forest types, dominant species, and mapping of functional guilds by hyperspectral and simulated multispectral Sentinel-2 data

Abstract To answer new scientific and ecological questions and monitor multiple forest changes, a fine scale characterization of these ecosystems is needed, and could imply the mapping of specific species, of detailed forest types, and of functional composition. This characterization can be now provided by the novel Earth Observation tools. This study aims to contribute to understanding the innovation in forest and ecological research that can be brought in by advanced remote sensing instruments, and proposes the guild mapping approach as a tool to efficiently monitor the varied tropical forest resources. We evaluated, in tropical Ghanaian forests, the ability of airborne hyperspectral and simulated multispectral Sentinel-2 data, and derived vegetation indices and textures, to: distinguish between two different forest types; to discriminate among selected dominant species; and to separate trees species grouped according to their functional guilds: Pioneer, Non Pioneer Light Demanding, and Shade Bearer. We then produced guild classification maps for each area using hyperspectral data. Our results showed that with both hyperspectral and simulated Sentinel-2 data these discrimination tasks can be successfully accomplished. Results also stressed the importance of texture features, especially if using the lower spectral and spatial Sentinel-2 resolution data, and highlighted the important role of the new Sentinel-2 data for ecological monitoring. Classification results showed a statistically significant improvement in overall accuracy using Support Vector Machine, over Maximum Likelihood approach. We proposed the functional guilds mapping as an innovative approach to: (i) monitor compositional changes, especially with respect to the effects of global climate change on forests, and particularly in the tropical biome where the occurrence of hundreds of species prevents mapping activities at species level; (ii) support large-scale forest inventories. The imminent Sentinel-2 data could serve to open the road for the development of new concepts and methods in forestry and ecological research.

Classification of Polarimetric SAR Images Based on Modeling Contextual Information and Using Texture Features

This paper proposes a novel contextual method for classification of polarimetric synthetic aperture radar data. The method combines support vector machine (SVM) and Wishart classifiers to benefit from both parametric and nonparametric methods. This method computes the energy function of a Markov random field (MRF) in the neighborhoods of the pixel using Wishart distribution. It then relates the Markovian energydifference function to the SVM classifier. Therefore, the salt-and-pepper effect on the classified map is reduced using a contextual classifier. Moreover, to achieve the full advantage of spatial information, texture features are added into the contextual classification. Texture features are extracted from SPAN images and are added to the SVM classifier. In this paper, two Radarsat-2 polarimetric images acquired in the leaf-off and leaf-on seasons are used from a forest area. Efficient multitemporal information is exploited using composite kernels in SVM. Comparison of the proposed algorithm with the Wishart, Wishart-MRF, SVM, and SVM with composite kernel classifiers shows a 21.72%, 16.17%, 11.29%, and 8.19% improvement in overall accuracy, respectively. Moreover, incorporating texture features into classification results significant increase in the average accuracy in forest species compared with the use of only polarimetric features.

DECISION LEVEL FUSION OF LIDAR DATA AND AERIAL COLOR IMAGERY BASED ON BAYESIAN THEORY FOR URBAN AREA CLASSIFICATION

Abstract. Airborne Light Detection and Ranging (LiDAR) generates high-density 3D point clouds to provide a comprehensive information from object surfaces. Combining this data with aerial/satellite imagery is quite promising for improving land cover classification. In this study, fusion of LiDAR data and aerial imagery based on Bayesian theory in a three-level fusion algorithm is presented. In the first level, pixel-level fusion, the proper descriptors for both LiDAR and image data are extracted. In the next level of fusion, feature-level, using extracted features the area are classified into six classes of “Buildings”, “Trees”, “Asphalt Roads”, “Concrete roads”, “Grass” and “Cars” using Naïve Bayes classification algorithm. This classification is performed in three different strategies: (1) using merely LiDAR data, (2) using merely image data, and (3) using all extracted features from LiDAR and image. The results of three classifiers are integrated in the last phase, decision level fusion, based on Naïve Bayes algorithm. To evaluate the proposed algorithm, a high resolution color orthophoto and LiDAR data over the urban areas of Zeebruges, Belgium were applied. Obtained results from the decision level fusion phase revealed an improvement in overall accuracy and kappa coefficient.

Structured prediction for differentiating between normal rhythms, ventricular tachycardia, and ventricular fibrillation in the ECG.

Recent studies have been performed on feature selection for diagnostics between non-ventricular rhythms and ventricular arrhythmias, or between non-ventricular fibrillation and ventricular fibrillation. However they did not assess classification directly between non-ventricular rhythms, ventricular tachycardia and ventricular fibrillation, which is important in both a clinical setting and preclinical drug discovery. In this study it is shown that in a direct multiclass setting, the selected features from these studies are not capable at differentiating between ventricular tachycardia and ventricular fibrillation. A high dimensional feature space, Fourier magnitude spectra, is proposed for classification, in combination with the structured prediction method conditional random fields. An improvement in overall accuracy, and sensitivity of every category under investigation is achieved.

Hyperspectral Image Denoising Using Low Pass Sparse Banded Filter Matrix for Improved Sparsity Based Classification

The recent advance in sensor technology is a boon for hyperspectral remote sensing. Though Hyperspectral images (HSI) are captured using these advanced sensors, they are highly prone to issues like noise, high dimensionality of data and spectral mixing. Among these, noise is the major challenge that affects the quality of the captured image. In order to overcome this issue, hyperspectral images are subjected to spatial preprocessing (denoising) prior to image analysis (Classification). In this paper, authors discuss a sparsity based denoising strategy which uses low pass sparse banded filter matrices (AB filter) to effectively denoise each band of HSI. Both subjective and objective evaluations are conducted to prove the efficiency of the proposed method. Subjective evaluations involve visual interpretation while objective evaluations deals with the computation of quality matrices such as Peak Signal to Noise Ratio (PSNR) and Structural Similarity (SSIM) index at different noise variance. In addition to these, the denoised image is followed by a sparsity based classification using Orthogonal Matching Pursuit (OMP) to evaluate the effect of various denoising techniques on classification. Classification indices obtained without and with applying preprocessing are compared to highlight the potential of the proposed method. The experiment is performed on standard Indian Pines Dataset. By using 10% of training set, a significant improvement in overall accuracy (84.21%) is obtained by the proposed method, compared to the other existing techniques.

Emergency physician awareness of prehospital procedures and medications.

IntroductionMaintaining patient safety during transition from prehospital to emergency department (ED) care depends on effective handoff communication between providers. We sought to determine emergency physicians’ (EP) knowledge of the care provided by paramedics in terms of both procedures and medications, and whether the use of a verbal report improved physician accuracy.MethodsWe conducted a 2-phase observational survey of a convenience sample of EPs in an urban, academic ED. In this large ED paramedics have no direct contact with physicians for non-critical patients, giving their report instead to the triage nurse. In Phase 1, paramedics gave verbal report to the triage nurse only. In Phase 2, a research assistant (RA) stationed in triage listened to this report and then repeated it back verbatim to the EPs caring for the patient. The RA then queried the EPs 90 minutes later regarding their patients’ prehospital procedures and medications. We compared the accuracy of these 2 reporting methods.ResultsThere were 163 surveys completed in Phase 1 and 116 in Phase 2. The oral report had no effect on EP awareness that the patient had been brought in by ambulance (86% in Phase 1 and 85% in Phase 2.) The oral report did improve EP awareness of prehospital procedures, from 16% in Phase 1 to 45% in Phase 2, OR=4.28 (2.5–7.5). EPs were able to correctly identify all oral medications in 18% of Phase 1 cases and 47% of Phase 2 cases, and all IV medications in 42% of Phase 1 cases and 50% of Phase 2 cases. The verbal report led to a mild improvement in physician awareness of oral medications given, OR=4.0 (1.09–14.5), and no improvement in physician awareness of IV medications given, OR=1.33 (0.15–11.35). Using a composite score of procedures plus oral plus IV medications, physicians had all three categories correct in 15% of Phase 1 and 39% of Phase 2 cases (p<0.0001).ConclusionEPs in our ED were unaware of many prehospital procedures and medications regardless of the method used to provide this information. The addition of a verbal hand-off report resulted in a modest improvement in overall accuracy.

Su2014 A Multichannel Wireless Mapping System for Monitoring Gastric Slow Wave Propagation

Aim To investigate the relative importance of both linear and circular polarization parameters, derived from 4x4 Mueller matrix measurements, in discriminating gastric cancer from normal gastric tissues over the visible spectral region from 470 nm to 632 nm. Methods A total of 46 tissue samples, obtained from 40 patients, were examined in this study, in which 26 normal gastric samples were obtained as endoscopic biopsies while 20 gastric cancer samples were obtained from gastrectomy. Informed consent was obtained before tissue samples were used in this study. For multiple normal samples from one single patient, the values of polarization parameters were averaged to represent only one sample in each classification. This yielded a total of 20 sets of data for normal samples and 20 sets of data for cancer samples, one set for each patient. Each sample was fixed using 10%-formalin solution before it was embedded in paraffin. Then two 4-μm vertical sections immediately next to each other were made in the sample. One tissue section was routinely stained with Hematoxylin and Eosin (H&E) to generate a pathological report. The other tissue section was not stained, which was placed on a microscope slide without a cover slip for polarimetry measurements. Results The studied polarization parameters were retardance, diattenuation and depolarization of normal and cancerous gastric samples at nine wavelengths from 470 nm to 632 nm. The average retardance of normal samples was smaller than that of cancer samples. There was no overlap in the error bars between normal samples and cancerous samples at all nine wavelengths, suggesting that retardance could be used to effectively differentiate normal and cancer samples. In contrast, there were significant overlaps in the error bars of diattenuation and depolarization between normal and cancer samples at all wavelengths. It was found that the combination of linear depolarization and linear retardance showed the highest overall accuracy (95.00%) among all the possible combinations of two parameters. This combination also demonstrated considerable improvement in overall accuracy compared to that for either linear retardance (82.50%) or linear depolarization (77.50%) alone. Conclusion There were significant differences in most polarization parameters between normal and cancer tissue samples. The combination of linear depolarization and linear retardance yielded the highest accuracy in sample classification.

Spectral normalization of SPOT 4 data to adjust for changing leaf phenology within seasonal forests in Cambodia

As cloud cover exacerbates the application of optical satellite data for forest monitoring in tropical wet and dry regions during the rainy season, data acquisition is mainly restricted to the dry season. When analyzing wide areas, large numbers of single scenes obtained at different times of the dry season are often handled. Such imagery is characterized by changes of spectral reflectance due to vegetation phenology, varying atmospheric effects and solar geometries. In order to allow batch processing with automatic classification techniques, inter-scene comparability is required and data have to be radiometrically normalized. Cambodia is characterized by a mixture of evergreen, semi-evergreen and deciduous forest types, the latter two experiencing at least partial leaf shedding over the course of the dry season. Using spatial medium resolution SPOT 4 data and a manually delineated base map a season adjustment model was developed. The model is adapting the land cover specific spectral signatures of a slave scene (acquired in the middle of the dry season with its seasonal forests defoliated) to an adjacent master scene (from the beginning of the dry season, showing the same forest types with leafs). The relative position of every pixel reflectance was determined in relation to the mean reflectance and its standard deviation for each land cover type and sensor band of the unadjusted slave scene. For seasonality adjustment these pixel reflectance values were transformed (rescaled) to the corresponding position in spectral space defined by the band mean reflectance and standard deviation derived from the corresponding land cover class of the master scene. While the variability of spectral profiles of the pixels in the slave scene is rescaled, the mean reflectance value of the land cover class in the slave scene is conformed to the mean reflectance of the corresponding land cover class in the master scene. The Transformed Divergence (TD) separability index was used to indicate the performance of the adjustment process by characterizing the spectral distance for each land cover type comparing a reference dataset to the uncorrected and to the seasonality corrected scene respectively. While the TD values of all forest types showed a sharp decline, highlighting the good performance of the model, the TD values of the agriculture/urban class remained high, indicating limited normalization of this heterogeneous land cover type. In order to further demonstrate the performance of the model, an object-based land cover classification was applied to the unadjusted as well as to the corresponding adjusted scene. A comparison of the results showed a highly significant improvement of overall accuracy from 32.2% to 75.8% when applying seasonality adjustment.

Image Training, Using Random Images of Melanoma, Performs as Well as the ABC(D) Criteria in Enabling Novices to Distinguish Between Melanoma and Mimics of Melanoma

Robust experimental evidence supporting many attempts to facilitate early melanoma diagnosis is lacking. In an experimental study using a browser interface we have examined diagnostic accuracy, sensitivity and specificity of novices in distinguishing between melanomas and mimics of melanoma. We show that rule-based ABC methods and image training, based on random images of melanoma, improve specificity to similar degrees, with-out effects on sensitivity, leading to small improvements in overall accuracy. There was a significant effect of age with older subjects performing better. Although both the ABC method and image training groups showed improved performance over the control group, overall performance was poor. For instance, for a task in which 1 in 4 test images was a melanoma, and 3 out of 4 benign, both interventions (ABC or image training) increased accuracy from the control value of 53% to around 61%. For reference, dermatology trainees performed at a much higher level of accuracy. Our study provides little support for the use of such methods in public education, but suggests ways in which performance might be improved.

Application traffic classification at the early stage by characterizing application rounds

This article proposes a machine learning-based high-accuracy algorithm called “APPlication Round method (APPR)” to identify network application traffic at the early stage. For each TCP/UDP flow, discriminators available at the early stage are determined to support high-accuracy and real-time traffic classification. Such discriminators characterize the possible negotiation behaviors of each flow from an application layer perspective. The ability of flow attributes is tested using several machine learning algorithms. By contrast, this study also compares the level of accuracy of the proposed method with those reported by machine learning-based application traffic classification methods that have addressed real-time application traffic classification problems based on identical sample traffic sets. By applying a pruned C4.5 tree machine learning algorithm to real traffic traces, the proposed method offers a maximal 99.21%, with an average overall accuracy of 92.88% for all traffic samples. Compared to other machine learning algorithms, the proposed algorithm not only provides a minimal accuracy improvement of approximately 7–8% for normal ratio data sets and more than 15–30% improvement of overall accuracy for fixed ratio data samples, but is also suitable for on-line identification because of the low-flow test duration. Furthermore, the proposed method is also appropriate for identifying encrypted protocols because it demonstrates high accuracy, classifies encryption-based protocols, and supports real-time classification.

Fast and efficient lung disease classification using hierarchical one-against-all support vector machine and cost-sensitive feature selection

To improve time and accuracy in differentiating diffuse interstitial lung disease for computer-aided quantification, we introduce a hierarchical support vector machine which selects a class by training a binary classifier at each node in a hierarchy, thus allowing each classifier to use a class-specific quasi-optimal feature set. In addition, the computational cost-sensitive group-feature selection criterion combined with the sequential forward selection is applied in order to obtain a useful and computationally inexpensive quasi-optimal feature set for the purpose of accelerating the classification time. The classification time was reduced by up to 57% and the overall accuracy was significantly improved in comparison with the one-against-all and one-against-one support vector machine methods with sequential forward selection (paired t-test, p<0.001). The reduction of classification time as well as the improvement of overall accuracy demonstrates promise for the proposed classification method to be adopted in various real-time and on-line image-based clinical applications.

SU-E-J-95: Towards Optimum Boundary Conditions for Biomechanical Model Based Deformable Registration Using Intensity Based Image Matching for Prostate Correlative Pathology.

Deformable registration of histology to MRI is an essential tool to validate in vivo prostate cancer imaging. However, direct registration of histology to in vivo MR is prone to error due to geometric differences between the tissue sections and the in vivo imaging planes. To increase the accuracy of registration, an ex vivo high resolution MRI is acquired to compensate for the direct registration difficulties. A novel intensity-based deformable registration algorithm based on local variation in image intensities is proposed to register the histology to ex vivo MRI of prostatectomy specimens. Four sets of ex vivo MR and whole mount pathology images from four patients were used to investigate and validate the technique. In addition, 9 synthetically deformed ex vivo MR images were used. The standard deviation in local windows within the images was calculated to generate intermediate images based on both MR and histology. The intermediate images were registered using the Drop package (Munich, Germany). To further increase the accuracy, a final refinement of the registration was performed using Drop with a finer B-spline rid. The registration parameters were tuned to achieve a visually acceptable registration. Magnitude of Differences (MOD) and Angular Error (AE) were used to validate the synthetic data, and the Target Registration Error (TRE) of manually indicated landmarks was used for the clinical data. MOD of 0.6mm and AE of 8.3 degrees showed the efficacy of using intermediate images, compared to 0.8mm and 10.0 degrees achieved with Drop without the intermediate images. The average mean±std TRE among the four patients was 1.0±0.6 mm using the proposed method compared to 1.6±1.1 mm using Elastix (Utrecht, The Netherlands). An intensity-based deformable registration algorithm which uses intermediate images was evaluated on prostatectomy specimens and synthetically deformed clinical data, indicating improvement in overall accuracy and robustness. OICR, Terry Fox Ultrasound for Cancer Therapy. Dr. Brock is a Cancer Care Ontario Research Chair in Cancer Imaging and has financial interest in deformable registration technology through the licensing of Morfeus to RaySearch Laboratories.