Search

Evidence is mounting that ConvNets are the best representation learning method for recognition. In the common scenario, a ConvNet is trained on a large labeled dataset and the feed-forward units activation, at a certain layer of the network, is used as a generic representation of an input image. Recent studies have shown this form of representation to be astoundingly effective for a wide range of recognition tasks. This paper thoroughly investigates the transferability of such representations w.r.t. several factors. It includes parameters for training the network such as its architecture and parameters of feature extraction. We further show that different visual recognition tasks can be categorically ordered based on their distance from the source task. We then show interesting results indicating a clear correlation between the performance of tasks and their distance from the source task conditioned on proposed factors. Furthermore, by optimizing these factors, we achieve stateof-the-art performances on 16 visual recognition tasks., QC 20150507. QC 20200701

This paper provides an extensive study on the availability of image representations based on convolutional networks (ConvNets) for the task of visual instance retrieval. Besides the choice of convolutional layers, we present an efficient pipeline exploiting multi-scale schemes to extract local features, in particular, by taking geometric invariance into explicit account, i.e. positions, scales and spatial consistency. In our experiments using five standard image retrieval datasets, we demonstrate that generic ConvNet image representations can outperform other state-of-the-art methods if they are extracted appropriately., QC 20200616

Discriminative latent variable models (LVM) are frequently applied to various visualrecognition tasks. In these systems the latent (hidden) variables provide a formalism formodeling structured variation of visual features. Conventionally, latent variables are de-fined on the variation of the foreground (positive) class. In this work we augment LVMsto includenegativelatent variables corresponding to the background class. We formalizethe scoring function of such a generalized LVM (GLVM). Then we discuss a frameworkfor learning a model based on the GLVM scoring function. We theoretically showcasehow some of the current visual recognition methods can benefit from this generalization.Finally, we experiment on a generalized form of Deformable Part Models with negativelatent variables and show significant improvements on two different detection tasks., QC 20150828

Supervised training of a convolutional network for object classification should make explicit any information related to the class of objects and disregard any auxiliary information associated with the capture of the image or thevariation within the object class. Does this happen in practice? Although this seems to pertain to the very final layers in the network, if we look at earlier layers we find that this is not the case. Surprisingly, strong spatial information is implicit. This paper addresses this, in particular, exploiting the image representation at the first fully connected layer,i.e. the global image descriptor which has been recently shown to be most effective in a range of visual recognition tasks. We empirically demonstrate evidences for the finding in the contexts of four different tasks: 2d landmark detection, 2d object keypoints prediction, estimation of the RGB values of input image, and recovery of semantic label of each pixel. We base our investigation on a simple framework with ridge rigression commonly across these tasks,and show results which all support our insight. Such spatial information can be used for computing correspondence of landmarks to a good accuracy, but should potentially be useful for improving the training of the convolutional nets for classification purposes., Qc 20150828

Syftet med offentlig upphandling är att konkurrensutsätta marknader och skapa bästa möjligaaffär för de offentliga medel som den upphandlande myndigheten har till sitt förfogande.Offentlig upphandling ska ske med långsiktigt perspektiv och det finns verktyg för att tillse attett långsiktigt synsätt erhålls. Ett av dessa verktyg är livscykelkostnader (LCC) som beaktaren produkts anskaffning, ägande och avveckling. En produkt som har ett högtanskaffningsvärde men låga framtida kostnader, kan bli det mest kostnadseffektiva inköpet.LCC används för att göra en ekonomisk värdering av olika investeringsalternativ,diskonterade över produktens livslängd. Livscykelkostnader är ett synsätt som utforskar deolika investeringsalternativen och används aktivt för att föra en dialog mellan aktörerna iupphandlingsprocessen. Användningen av livscykelkostnader, kan antas vara till god hjälp för beslutsfattarna i vattenoch avloppsbranschen (VA) om de vill ta långsiktigt korrekta beslut. Problematiken är attingen har kartlagt hur ett livscykelkostnadsverktyg påverkar beslutsprocessen i offentligaupphandlingar inom VA-branschen och om det ger beställaren ett tydligt beslutsstöd. Dennastudie syftade därför att ta reda på hur livscykelkostnader används i praktiken och hur enanalys av livscykelkostnader bättre kan stödja beslutsprocessen för offentliga upphandlingarinom VA. Studien syftade även till att ta reda på hur ett livscykelkostnadsverktyg bör varautformat, genom att ge förslag på förutsättningar som måste tas hänsyn till. Detta examensarbete har utförts på uppdrag av Vattenreningsindustrins mötesplats (VARIM).En kvalitativ intervjustudie genomfördes, uppdelad i två faser, med 27 stycken intervjuer somgav upphov till ett rikt material. Intervjupersonerna hade fem olika aktörsroller: beställare,upphandlare, brukare, konsult och anbudsgivare. Studien avgränsades till att endast undersökafyra kommuntyper: storstäder, större städer, varuproducerande kommuner ochglesbygdskommuner. Resultat, The purpose of public procurement is to create competitive markets and generate the bestpossible deal for the contracting authority. Public procurement should aim for long-termperspective, and there are tools to ensure that a long-term view is obtained. One tool is lifecycle costs (LCC), where product acquisition, ownership and settlement are considered. Aproduct that has a high investment cost but low operating costs, may be the most costeffectivepurchase. LCC is used to achieve an economic evaluation of different investmentoptions, discounted over the lifetime of the product. Life-cycle costing is an approach thatexplores the various investment options and is used to as a basis of discussion by theparticipants in the procurement process. The use of life-cycle costs, are expected to be useful for decision makers in the water andwastewater industry if they want to take long-term correct decisions. The problem is that noone has studied how a life cycle cost tool affects the decision-making process in publicprocurement within the water and wastewater industry and if it gives the client a gooddecision making basis. This study therefore aimed to find out how life cycle costs are used inpractice and how these can be used in practice, as an aid in the decision making process forwater and wastewater procurements. The study also aimed to find out how a life cycle costtool should be designed, by giving suggestions on conditions that must be taken into account. This thesis has been carried out on behalf of The meeting place for the industry in watertreatment (VARIM), with the aim to investigate the procurement process within the water andwastewater trade. It also studies how life cycle costs are used in the process. The study wascarried out in two steps, consisting of qualitative interviews, 27 individual interviews gaverise to a rich material, which was subject to further analysis. The interviewees had fivedifferent roles in the procurement process: client, purchaser, user

Publicerad på Lakartidningen.se 2015-04-20

Purpose: We investigated hospital readmission frequency during the 90 days after radical prostatectomy and assessed the readmission risk associated with potentially related variables. Materials and Methods: Using the population based, nationwide PCBaSe (Prostate Cancer data Base Sweden) we identified men diagnosed with incident prostate cancer between 2000 and 2011 who underwent radical prostatectomy as primary treatment. We used logistic regression analysis to examine the association of the risk of 90-day postoperative readmission with surgical method, calendar period, tumor risk category, hospital case load and patient characteristics. Results: During 90 postoperative days 2,317 of the 24,122 men (10%) identified were nonelectively readmitted, specifically 10% after retropubic, 9% after robot-assisted and 11% after laparoscopic radical prostatectomy. The range in readmission frequency among hospitals was 0% to 35%. Higher readmission risk was associated with the early calendar period (2009 to 2011 vs 2000 to 2002 OR 0.71, 95% CI 0.61-0.83), greater age (70 or greater vs less than 60 years OR 1.17, 95% CI 1.00-1.36), higher risk category (high vs low OR 1.78, 95% CI 1.57-2.03), high comorbidity (Charlson comorbidity index 3 or greater vs 0 OR 1.77, 95% CI 1.29-2.44) and low hospital surgical volume (150 or greater vs fewer than 30 radical prostatectomies per year OR 0.70, 95% CI 0.60-0.81). Conclusions: Readmission rates after different radical prostatectomy methods were similar, ranging from 9% to 11%, with wide variation among hospitals. Readmission rates can be used as an indicator of perioperative care quality but potential confounders must be adjusted to avoid bias.

Purpose: Transrectal ultrasound guided biopsy is the gold standard for detecting prostate cancer but international reports suggest that increasing risks are associated with the procedure. We estimated incidence and risk factors for infection after prostate biopsy as well as 90-day mortality using a nationwide Swedish sample. Material and Methods: We performed a population based study of 51,321 men from PCBaSe between 2006 and 2011. Primary outcome measures were dispensed prescriptions of antibiotics for urinary tract infection and hospitalization with a discharge diagnosis of urinary tract infection. Multivariable logistic regression was used to examine risk factors for infection in men who underwent prostate biopsy. Results: During the 6 months before biopsy the background incidence of urinary tract infection was approximately 2%. Within 30 days after biopsy 6% of the men had a dispensed prescription for urinary tract antibiotics and 1% were hospitalized with infection. The strongest risk factors for an antibiotic prescription were prior infection (OR 1.59, 95% CI 1.45-1.73), high Charlson comorbidity index (OR 1.25, 95% CI 1.11-1.41) and diabetes (OR 1.32, 95% CI 1.17-1.49). Risk of an antibiotic prescription after biopsy decreased from 2006 to 2011 (OR 0.79, 95% CI 0.70-0.90) but the risk of hospital admission increased (OR 2.14, 95% CI 1.58-2.94). No significant increase was observed in 90-day mortality. Conclusions: Severe infections with hospitalization after prostate biopsy are increasing in Sweden. The risk of post-biopsy infection is highest in men with a history of urinary tract infection and those with significant comorbidities.

Recent results indicate that the generic descriptors extracted from the convolutional neural networks are very powerful. This paper adds to the mounting evidence that this is indeed the case. We report on a series of experiments conducted for different recognition tasks using the publicly available code and model of the OverFeat network which was trained to perform object classification on ILSVRC13. We use features extracted from the OverFeat network as a generic image representation to tackle the diverse range of recognition tasks of object image classification, scene recognition, fine grained recognition, attribute detection and image retrieval applied to a diverse set of datasets. We selected these tasks and datasets as they gradually move further away from the original task and data the OverFeat network was trained to solve. Astonishingly, we report consistent superior results compared to the highly tuned state-of-the-art systems in all the visual classification tasks on various datasets. For instance retrieval it consistently outperforms low memory footprint methods except for sculptures dataset. The results are achieved using a linear SVM classifier (or L2 distance in case of retrieval) applied to a feature representation of size 4096 extracted from a layer in the net. The representations are further modified using simple augmentation techniques e.g. jittering. The results strongly suggest that features obtained from deep learning with convolutional nets should be the primary candidate in most visual recognition tasks., Best Paper Runner-up Award.QC 20140825

In this paper, we discuss the properties of a class of latent variable models that assumes each labeled sample is associated with set of different features, with no prior knowledge of which feature is the most relevant feature to be used. Deformable-Part Models (DPM) can be seen as good example of such models. While Latent SVM framework (LSVM) has proven to be an efficient tool for solving these models, we will argue that the solution found by this tool is very sensitive to the initialization. To decrease this dependency, we propose a novel clustering procedure, for these problems, to find cluster centers that are shared by several sample sets while ignoring the rest of the cluster centers. As we will show, these cluster centers will provide a robust initialization for the LSVM framework., QC 20150610

We discuss the properties of a class of latent variable models that assumes each labeled sample is associated with a set of different features, with no prior knowledge of which feature is the most relevant feature to be used. Deformable-Part Models (DPM) can be seen as good examples of such models. These models are usually considered to be expensive to train and very sensitive to the initialization. In this paper, we focus on the learning of such models by introducing a topological framework and show how it is possible to both reduce the learning complexity and produce more robust decision boundaries. We will also argue how our framework can be used for producing robust decision boundaries without exploiting the dataset bias or relying on accurate annotations. To experimentally evaluate our method and compare with previously published frameworks, we focus on the problem of image classification with object localization. In this problem, the correct location of the objects is unknown, during both training and testing stages, and is considered as a latent variable., QC 20150605

This paper demonstrates a system for automatic detection of visual attention and identification of salient items at exhibitions (e.g. museum or an auction). The method is offline and is done on a video captured by a head mounted camera. Towards the estimation of attention, we define the notions of "saliency" and "interestingness" for an exhibition items. Our method is a combination of multiple state of the art techniques from different vision tasks such as tracking, image matching and retrieval. Many experiments are conducted to evaluate multiple aspects of our method. The method has proven to be robust to image blur, occlusion, truncation, and dimness. The experiments shows strong performance for the tasks of matching items, estimating focus frames and detecting salient and interesting items. This can be useful to the commercial vendors and museum curators and help them to understand which items are appealing more to the visitors., QC 20150521

In this paper, we propose a framework for gradually improving the quality of an already existing image descriptor. The descriptor used in this paper (Afkham et al., 2013) uses the response of a series of discriminative components for summarizing each image. As we will show, this descriptor has an ideal form in which all categories become linearly separable. While, reaching this form is not feasible, we will argue how by replacing a small fraction of these components, it is possible to obtain a descriptor which is, on average, closer to this ideal form. To do so, we initially identify which components do not contribute to the quality of the descriptor and replace them with more robust components. Here, a joint feature selection method is used to find improved components. As our experiments show, this change directly reflects in the capability of the resulting descriptor in discriminating between different categories., QC 20150615

PURPOSE: To investigate readmission frequencies during the 90 days following radical prostatectomy and to assess readmission risk associated with potentially related variables. MATERIALS AND METHODS: Using the population-based, nationwide database Prostate Cancer data Base Sweden (PCBaSe), we identified men diagnosed with incident prostate cancer between 2000 and 2011 who underwent radical prostatectomy (RP) as their primary treatment, and we used logistic regression analysis to examine the association of the risk of 90-day postoperative readmission with surgical method, calendar period, tumor risk category, hospital case load, and patient characteristics. RESULTS: During the 90 postoperative days, 2,317 (10%) of the 24,122 men identified were non-electively readmitted, specifically 10% after retropubic radical prostatectomy (RRP), 9% after robot-assisted RP (RALP) and 11% after laparoscopic RP (LRP). The range in the readmission frequency between hospitals was 0-35%. A higher risk of readmission was associated with early calendar period (2009-2011 vs. 2000-2002: odds ratio (OR), 0.71; 95% confidence interval (CI), 0.61-0.83), greater age (≥70 years vs. <60 years: OR, 1.17; 95% CI, 1.00-1.36), higher risk category (high vs. low-risk category: OR, 1.78; 95% CI, 1.57-2.03), high comorbidity (Charlson comorbidity index ≥3 vs. 0: OR, 1.77; 95% CI, 1.29-2.44), and low hospital surgical volume (≥150 vs. <30 RPs per year: OR, 0.70; 95% CI, 0.60-0.81). CONCLUSIONS: Readmission rates after different RP methods were similar, ranging from 9% to 11%, with a wide variation between hospitals. Readmission rates can be used as an indicator of perioperative care quality, but potential confounders need to be adjusted to avoid bias.

Background: Men with germline breast cancer 1, early onset (BRCA1) or breast cancer 2, early onset (BRCA2) gene mutations have a higher risk of developing prostate cancer (PCa) than noncarriers. IMPACT (Identification of Men with a genetic predisposition to ProstAte Cancer: Targeted screening in BRCA1/2 mutation carriers and controls) is an international consortium of 62 centres in 20 countries evaluating the use of targeted PCa screening in men with BRCA1/2 mutations. Objective: To report the first year's screening results for all men at enrolment in the study. Design, setting and participants: We recruited men aged 40-69 yr with germline BRCA1/2 mutations and a control group of men who have tested negative for a pathogenic BRCA1 or BRCA2 mutation known to be present in their families. All men underwent prostate-specific antigen (PSA) testing at enrolment, and those men with PSA > 3 ng/ml were offered prostate biopsy. Outcome measurements and statistical analysis: PSA levels, PCa incidence, and tumour characteristics were evaluated. The Fisher exact test was used to compare the number of PCa cases among groups and the differences among disease types. Results and limitations: We recruited 2481 men (791 BRCA1 carriers, 531 BRCA1 controls; 731 BRCA2 carriers, 428 BRCA2 controls). A total of 199 men (8%) presented with PSA > 3.0 ng/ml, 162 biopsies were performed, and 59 PCas were diagnosed (18 BRCA1 carriers, 10 BRCA1 controls; 24 BRCA2 carriers, 7 BRCA2 controls); 66% of the tumours were classified as intermediate-or high-risk disease. The positive predictive value (PPV) for biopsy using a PSA threshold of 3.0 ng/ml in BRCA2 mutation carriers was 48%-double the PPV reported in population screening studies. A significant difference in detecting intermediate-or high-risk disease was observed in BRCA2 carriers. Ninety-five percent of the men were white, thus the results cannot be generalised to all ethnic groups. Conclusions: The IMPACT screening network will be useful

Projecte realitzat en el marc d’un programa de mobilitat amb la KTH Royal Institute of Technology, [ANGLÈS] Face recognition is a fundamental task in computer vision and has been an important field of study for many years. This report presents a unified process for automatically extract a set of face landmarks and remove all differences related to pose, expression and environment by bringing faces to a neutral pose-centered state. Landmark detection is based on a multiple viewpoint Pictorial Structure model. In this project we address both the problem of how to find a set of landmarks from a model and the problem of training such a model from a set of labelled examples. We show how such a model successfully captures a great range of deformations needing far less training examples than common commercial face detectors. The alignment process basically aims to remove differences between multiple faces so they all can be analysed under the same criteria. It is carried out with Thin-plate Splines to adjust the detected set of landmarks to the desired configuration. We present results of our algorithms both in a constrained environment and in the challenging LFPW face database. Successful outcomes are shown that prove our method to be a solid process for unitedly recognise and warp faces in the wild and to be on a par with other state-of-the-art procedures., [CASTELLÀ] El reconocimiento y clasificación de caras representa uno de los campos de estudio más importantes dentro la visión artificial por computadora y el procesado de imagen. En este proyecto se presenta un proceso unificado que extrae automáticamente, a partir de imágenes, un conjunto de puntos faciales característicos, eliminando al mismo tiempo las diferencias relacionadas con la orientación, expresión e iluminación. En una primera fase, se utiliza un modelo de Estructuras Pictográficas Multivista para la detección de puntos característicos. En este informe, se atienden con detalle tanto el problema de extraer puntos con este modelo, como el proceso de aprendizaje a partir de imágenes anotadas. En una segunda fase, se eliminan las diferencias entre caras mediante el uso de Thin-Plate Splines, para que todas las imágenes puedan ser analizadas bajo el mismo criterio en posibles aplicaciones futuras. Los algoritmos implementados, representan hasta la fecha, el primer sistema desarrollado, que simultáneamente extrae y alinea caras, y demuestran un comportamiento muy fiable y preciso tanto en entornos controlados como libres (base de datos LFPW)., [CATALÀ] reconeixement i classificació de cares és un dels camps d'estudi més importants dintre la visió artificial per computadora i el processament d'imatge. En aquest projecte es presenta un procés unificat que extreu automàticament, a partir d'imatges, un conjunt de punts facials característics eliminant al mateix temps les diferències relacionades amb l'orientació, expressió i il·luminació. En una primera fase, s'utilitza un model d'Estructures Pictogràfiques Multivista per a la detecció de punts característics. En aquest informe, s'adreça amb detall tant el problema d'extreure punts amb aquest model com el procés d'aprenentatge a partir d'imatges anotades. En una segona fase, s'eliminen les diferències entre cares mitjançant Thin-Plate Splines, per tal que totes les imatges puguin ser analitzades sota el mateix criteri en possibles aplicacions futures. Els algoritmes implementats, representen fins al moment, el primer sistema desenvolupat que unificadament extreu i alinea cares, i demostren un comportament molt fiable i acurat tant en entorns controlats com lliures (base de dades LFPW).

Supervised training of a convolutional network for object classification should make explicit any information related to the class of objects and disregard any auxiliary information associated with the capture of the image or the variation within the object class. Does this happen in practice? Although this seems to pertain to the very final layers in the network, if we look at earlier layers we find that this is not the case. Surprisingly, strong spatial information is implicit. This paper addresses this, in particular, exploiting the image representation at the first fully connected layer, i.e. the global image descriptor which has been recently shown to be most effective in a range of visual recognition tasks. We empirically demonstrate evidences for the finding in the contexts of four different tasks: 2d landmark detection, 2d object keypoints prediction, estimation of the RGB values of input image, and recovery of semantic label of each pixel. We base our investigation on a simple framework with ridge rigression commonly across these tasks, and show results which all support our insight. Such spatial information can be used for computing correspondence of landmarks to a good accuracy, but should potentially be useful for improving the training of the convolutional nets for classification purposes.

Computer vision tasks are traditionally defined and evaluated using semantic categories. However, it is known to the field that semantic classes do not necessarily correspond to a unique visual class (e.g. inside and outside of a car). Furthermore, many of the feasible learning techniques at hand cannot model a visual class which appears consistent to the human eye. These problems have motivated the use of 1) Unsupervised or supervised clustering as a preprocessing step to identify the visual subclasses to be used in a mixture-of-experts learning regime. 2) Felzenszwalb et al. part model and other works model mixture assignment with latent variables which is optimized during learning 3) Highly non-linear classifiers which are inherently capable of modelling multi-modal input space but are inefficient at the test time. In this work, we promote an incremental view over the recognition of semantic classes with varied appearances. We propose an optimization technique which incrementally finds maximal visual subclasses in a regularized risk minimization framework. Our proposed approach unifies the clustering and classification steps in a single algorithm. The importance of this approach is its compliance with the classification via the fact that it does not need to know about the number of clusters, the representation and similarity measures used in pre-processing clustering methods a priori. Following this approach we show both qualitatively and quantitatively significant results. We show that the visual subclasses demonstrate a long tail distribution. Finally, we show that state of the art object detection methods (e.g. DPM) are unable to use the tails of this distribution comprising 50\% of the training samples. In fact we show that DPM performance slightly increases on average by the removal of this half of the data., Comment: Updated ICCV 2013 submission

Recent results indicate that the generic descriptors extracted from the convolutional neural networks are very powerful. This paper adds to the mounting evidence that this is indeed the case. We report on a series of experiments conducted for different recognition tasks using the publicly available code and model of the \overfeat network which was trained to perform object classification on ILSVRC13. We use features extracted from the \overfeat network as a generic image representation to tackle the diverse range of recognition tasks of object image classification, scene recognition, fine grained recognition, attribute detection and image retrieval applied to a diverse set of datasets. We selected these tasks and datasets as they gradually move further away from the original task and data the \overfeat network was trained to solve. Astonishingly, we report consistent superior results compared to the highly tuned state-of-the-art systems in all the visual classification tasks on various datasets. For instance retrieval it consistently outperforms low memory footprint methods except for sculptures dataset. The results are achieved using a linear SVM classifier (or $L2$ distance in case of retrieval) applied to a feature representation of size 4096 extracted from a layer in the net. The representations are further modified using simple augmentation techniques e.g. jittering. The results strongly suggest that features obtained from deep learning with convolutional nets should be the primary candidate in most visual recognition tasks., Comment: version 3 revisions: 1)Added results using feature processing and data augmentation 2)Referring to most recent efforts of using CNN for different visual recognition tasks 3) updated text/caption

Evidence is mounting that Convolutional Networks (ConvNets) are the most effective representation learning method for visual recognition tasks. In the common scenario, a ConvNet is trained on a large labeled dataset (source) and the feed-forward units activation of the trained network, at a certain layer of the network, is used as a generic representation of an input image for a task with relatively smaller training set (target). Recent studies have shown this form of representation transfer to be suitable for a wide range of target visual recognition tasks. This paper introduces and investigates several factors affecting the transferability of such representations. It includes parameters for training of the source ConvNet such as its architecture, distribution of the training data, etc. and also the parameters of feature extraction such as layer of the trained ConvNet, dimensionality reduction, etc. Then, by optimizing these factors, we show that significant improvements can be achieved on various (17) visual recognition tasks. We further show that these visual recognition tasks can be categorically ordered based on their distance from the source task such that a correlation between the performance of tasks and their distance from the source task w.r.t. the proposed factors is observed., Comment: Extended version of the workshop paper with more experiments and updated text and title. Original CVPR15 DeepVision workshop paper title: "From Generic to Specific Deep Representations for Visual Recognition"

Objective. Only a minority of patients with low-risk prostate cancer needs treatment, but the methods for optimal selection of patients for treatment are not established. This article describes the Study of Active Monitoring in Sweden (SAMS), which aims to improve those methods. Material and methods. SAMS is a prospective, multicentre study of active surveillance for low-risk prostate cancer. It consists of a randomized part comparing standard rebiopsy and follow-up with an extensive initial rebiopsy coupled with less intensive follow-up and no further scheduled biopsies (SAMS-FU), as well as an observational part (SAMS-ObsQoL). Quality of life is assessed with questionnaires and compared with patients receiving primary curative treatment. SAMS-FU is planned to randomize 500 patients and SAMS-ObsQoL to include at least 500 patients during 5 years. The primary endpoint is conversion to active treatment. The secondary endpoints include symptoms, distant metastases and mortality. All patients will be followed for 10-15 years. Results. Inclusion started in October 2011. In March 2013, 148 patients were included at 13 Swedish urological centres. Conclusions. It is hoped that the results of SAMS will contribute to fewer patients with indolent, low-risk prostate cancer receiving unnecessary treatment and more patients on active surveillance who need treatment receiving it when the disease is still curable. The less intensive investigational follow-up in the SAMS-FU trial would reduce the healthcare resources allocated to this large group of patients if it replaced the present standard schedule.

BACKGROUND: Prostate cancer (PCa) and surgery are both associated with increased risk of thromboembolic diseases (TED). OBJECTIVE: We assessed risk of TED among men undergoing different types of urologic surgery. DESIGN, SETTING, AND PARTICIPANTS: Using the Prostate Cancer Database Sweden (PCBaSe) Sweden, we identified all men (n=45 065) undergoing pelvic lymph node dissection (PLND), radical prostatectomy (RP) with or without PLND, orchiectomy due to PCa, or a transurethral resection of the prostate (TURP). We identified a comparison cohort from the population. OUTCOME MEASUREMENTS AND STATISTICAL ANALYSIS: Main outcomes were deep venous thrombosis (DVT) and pulmonary embolism (PE) as primary diagnoses in the National Patient Register or Cause of Death Register (2002-2010). We calculated hazard ratios (HR) and 95% confidence intervals (CI) using multivariable Cox proportional hazards models. RESULTS AND LIMITATIONS: All surgical procedures were associated with increased risk of TED; laparoscopic and open RP with a PLND were the most strongly associated with TED (HR for PE: 8.1 [95% CI, 2.9-23.0] and 7.8 [95% CI, 4.9-13], respectively). For surgery including a PLND, the risk increased during the second half of the first postoperative month. The HR for PE after TURP in men with PCa was 3.0 (95% CI, 1.8-5.1). Patients with a history of TED had a strongly increased risk of TED (HR for DVT: 4.5; 95% CI, 2.6-8.0). A limitation is lack of information on TED prophylaxis, but its use was standardized during the study period for RP and PLND. Other limitations are lack of information on extent of PLND and lifestyle factors. CONCLUSIONS: Surgeries for PCa, including TURP, are associated with hospitalization for TED. Patients with a history of TED and patients undergoing a PLND were at highest risk. The largest risk was observed from days 14 to 28 postoperatively. Thus, our results suggest that prophylactic measures may be beneficial during the first 4 wk in these patients.

Creating a single feature descriptors from a collection of feature responses is an often occurring task. As such the bag-of-words descriptors have been very successful and applied to data from a large range of different domains. Central to this approach is making an association of features to words. In this paper we present a new and novel approach to feature to word association problem. The proposed method creates a more robust representation when data is noisy and requires less words compared to the traditional methods while retaining similar performance. We experimentally evaluate the method on a challenging image classification data-set and show significant improvement to the state of the art., QC 20131204