We propose a new framework for image classification with deep neural networks. The framework introduces intermediate outputs to the computational graph of a network. This enables flexible control of the computational load and balances the tradeoff between accuracy and execution time. Moreover, we present an interesting finding that the intermediate outputs can act as a regularizer at training time, improving the prediction accuracy. In the experimental section we demonstrate the performance of our proposed framework with various commonly used pretrained deep networks in the use case of apparent age estimation.

In this paper we study the real time deployment of deep learning algorithms in low resource computational environments. As the use case, we compare the accuracy and speed of neural networks for smile detection using different neural network architectures and their system level implementation on NVidia Jetson embedded platform. We also propose an asynchronous multithreading scheme for parallelizing the pipeline. Within this framework, we experimentally compare thirteen widely used network topologies. The experiments show that low complexity architectures can achieve almost equal performance as larger ones, with a fraction of computation required.

Glandular epithelial cells differentiate into complex multicellular or acinar structures, when embedded in three-dimensional (3D) extracellular matrix. The spectrum of different multicellular morphologies formed in 3D is a sensitive indicator for the differentiation potential of normal, non-transformed cells compared to different stages of malignant progression. In addition, single cells or cell aggregates may actively invade the matrix, utilizing epithelial, mesenchymal or mixed modes of motility. Dynamic phenotypic changes involved in 3D tumor cell invasion are sensitive to specific small-molecule inhibitors that target the actin cytoskeleton. We have used a panel of inhibitors to demonstrate the power of automated image analysis as a phenotypic or morphometric readout in cell-based assays. We introduce a streamlined stand-alone software solution that supports large-scale high-content screens, based on complex and organotypic cultures. AMIDA (Automated Morphometric Image Data Analysis) allows quantitative measurements of large numbers of images and structures, with a multitude of different spheroid shapes, sizes, and textures. AMIDA supports an automated workflow, and can be combined with quality control and statistical tools for data interpretation and visualization. We have used a representative panel of 12 prostate and breast cancer lines that display a broad spectrum of different spheroid morphologies and modes of invasion, challenged by a library of 19 direct or indirect modulators of the actin cytoskeleton which induce systematic changes in spheroid morphology and differentiation versus invasion. These results were independently validated by 2D proliferation, apoptosis and cell motility assays. We identified three drugs that primarily attenuated the invasion and formation of invasive processes in 3D, without affecting proliferation or apoptosis. Two of these compounds block Rac signalling, one affects cellular cAMP/cGMP accumulation. Our approach supports the growing needs for user-friendly, straightforward solutions that facilitate large-scale, cell-based 3D assays in basic research, drug discovery, and target validation.

BACKGROUND:Work-related stress is a significant problem for both people and organizations. It may lead to mental illnesses such as anxiety and depression, resulting in increased work absences and disabilities. Scalable interventions to prevent and manage harmful stress can be delivered with the help of technology tools to support self-observations and skills training.OBJECTIVE:The aim of this study was to assess the feasibility of the P4Well intervention in treatment of stress-related psychological problems. P4Well is a novel intervention which combines modern psychotherapy (the cognitive behavioral therapy and the acceptance and commitment therapy) with personal health technologies to deliver the intervention via multiple channels, includinggroup meetings, Internet/Web portal, mobile phone applications, and personal monitoring devices.METHODS:This pilot study design was a small-scale randomized controlled trial that compared the P4Well intervention with a waiting list control group. In addition to personal health technologies for self-assessment, the intervention consisted of 3 psychologist-assisted group meetings. Self-assessed psychological measures through questionnaires were collected offline pre- and post-intervention, and 6 months after the intervention for the intervention group. Acceptance and usage of technology tools were measured with user experience questionnaires and usage logs.RESULTS:A total of 24 subjects were randomized: 11 participants were followed up in the intervention group (1 was lost to follow-up) and 12 participants did not receive any intervention (control group). Depressive and psychological symptoms decreased and self-rated health and working ability increased. All participants reported they had benefited from the intervention. All technology tools had active users and 10/11 participants used at least 1 tool actively. Physiological measurements with personal feedback were considered the most useful intervention component.CONCLUSIONS:Our results confirm the feasibility of the intervention and suggest that it had positive effects on psychological symptoms, self-rated health, and self-rated working ability. The intervention seemed to have a positive impact on certain aspects of burnout and job strain, such as cynicism and over-commitment. Future studies need to investigate the effectiveness, benefits, and possible problems of psychological interventions which incorporate new technologies.TRIAL REGISTRATION:The Finnish Funding Agency for Technology and Innovation (TEKES), Project number 40011/08.

Normal prostate and some malignant prostate cancer (PrCa) cell lines undergo acinar differentiation and form spheroids in three-dimensional (3-D) organotypic culture. Acini formed by PC-3 and PC-3M, less pronounced also in other PrCa cell lines, spontaneously undergo an invasive switch, leading to the disintegration of epithelial structures and the basal lamina, and formation of invadopodia. This demonstrates the highly dynamic nature of epithelial plasticity, balancing epithelial-to-mesenchymal transition against metastable acinar differentiation. This study assessed the role of lipid metabolites on epithelial maturation. PC-3 cells completely failed to form acinar structures in delipidated serum. Adding back lysophosphatidic acid (LPA) and sphingosine-1-phosphate (S1P) rescued acinar morphogenesis and repressed invasion effectively. Blocking LPA receptor 1 (LPAR1) functions by siRNA (small interference RNA) or the specific LPAR1 inhibitor Ki16425 promoted invasion, while silencing of other G-protein-coupled receptors responsive to LPA or S1P mainly caused growth arrest or had no effects. The G-proteins Gα12/13 and Gαi were identified as key mediators of LPA signalling via stimulation of RhoA and Rho kinases ROCK1 and 2, activating Rac1, while inhibition of adenylate cyclase and accumulation of cAMP may be secondary. Interfering with these pathways specifically impeded epithelial polarization in transformed cells. In contrast, blocking the same pathways in non-transformed, normal cells promoted differentiation. We conclude that LPA and LPAR1 effectively promote epithelial maturation and block invasion of PrCa cells in 3-D culture. The analysis of clinical transcriptome data confirmed reduced expression of LPAR1 in a subset of PrCa's. Our study demonstrates a metastasis-suppressor function for LPAR1 and Gα12/13 signalling, regulating cell motility and invasion versus epithelial maturation.

Prostate epithelial cells from both normal and cancer tissues, grown in three-dimensional (3D) culture as spheroids, represent promising in vitro models for the study of normal and cancer-relevant patterns of epithelial differentiation. We have developed the most comprehensive panel of miniaturized prostate cell culture models in 3D to date (n = 29), including many non-transformed and most currently available classic prostate cancer (PrCa) cell lines. The purpose of this study was to analyze morphogenetic properties of PrCa models in 3D, to compare phenotypes, gene expression and metabolism between 2D and 3D cultures, and to evaluate their relevance for pre-clinical drug discovery, disease modeling and basic research. Primary and non-transformed prostate epithelial cells, but also several PrCa lines, formed well-differentiated round spheroids. These showed strong cell-cell contacts, epithelial polarization, a hollow lumen and were covered by a complete basal lamina (BL). Most PrCa lines, however, formed large, poorly differentiated spheroids, or aggressively invading structures. In PC-3 and PC-3M cells, well-differentiated spheroids formed, which were then spontaneously transformed into highly invasive cells. These cell lines may have previously undergone an epithelial-to-mesenchymal transition (EMT), which is temporarily suppressed in favor of epithelial maturation by signals from the extracellular matrix (ECM). The induction of lipid and steroid metabolism, epigenetic reprogramming, and ECM remodeling represents a general adaptation to 3D culture, regardless of transformation and phenotype. In contrast, PI3-Kinase, AKT, STAT/interferon and integrin signaling pathways were particularly activated in invasive cells. Specific small molecule inhibitors targeted against PI3-Kinase blocked invasive cell growth more effectively in 3D than in 2D monolayer culture, or the growth of normal cells. Our panel of cell models, spanning a wide spectrum of phenotypic plasticity, supports the investigation of different modes of cell migration and tumor morphologies, and will be useful for predictive testing of anti-cancer and anti-metastatic compounds.

Chronic health problems related to mental wellbeing are rapidly growing, calling for novel solutions focusing on individual as a psychophysiological being. We describe a novel technology-based concept for empowering citizen towards holistic self-management of her wellbeing: “P4Well” (Pervasive Personal and PsychoPhysiological management of WELLness). The primary focus of the concept is on management of stress and recovery from stress caused by daily life through improved health management strategies. The P4Well concept combines modern psychological methods with personal health technologies. The technologies include a web-portal and web-based tools, mobile phone with mobile client applications, wearable health monitoring devices, and different analysis methods based on physiological models for interpretation and feedback. The concept supports secured private expert consultation and peer-support through social media. Our driving principle is to recognize that an individual is the best master of her own wellness, and we target to empower her for realizing the fact.

In prevention of chronic diseases, health promotion and early interventions based on self-management should be emphasized. Mental health problems and stress cause a significant portion of healthcare costs, and also complicate the management of other chronic conditions. In addition to physical health, psychophysiological and social wellbeing should be equally promoted. Thus, we have previously designed and reported the P4Well or pervasive personal and psychophysiological management of wellness concept for working-age citizens. The concept supports the stress and recovery management on a daily basis through improved health management strategies, and combines psychological methods with personal health technologies. In this paper, we discuss the preliminary user study experiences of ongoing evaluations with two different user groups consisting of: 1) middle-aged men who are using the concept for managing their mental wellbeing or mild depression; and 2) entrepreneurs who are using the concept for coping with stress. Our results provide a preliminary assessment of the role and importance of experts, technologies, and peer-support in the concept.

Single photon emission computed tomography (SPECT) provides a method to obtain three-dimensional images of blood perfusion. In cardiac stress-rest SPECT studies, coronary artery disease can be examined by comparing intensity differences of the acquired images of the heart at stress and rest. The stress-rest image pair needs to be aligned into the same spatial position and orientation in order to compare the images in an automated and reproducible manner. In this work, we study a novel stackgram-based alignment algorithm for cardiac stress-rest SPECT data. The stackgram approach allows the alignment of the acquired raw data prior to image reconstruction, unlike conventional alignment techniques.

In low-dose X-ray computed tomography (CT), it is necessary to reduce the excessive X-ray quantum noise of the acquired data. This is the case especially when the filtered back-projection method is used for image reconstruction from sinogram data. In this study, we compare the conventional radial filtering direction of the sinogram to the recently introduced angular stackgram filtering technique. Our experimental tests were performed on simulated low-dose CT data. We modelled the X-ray quantum noise using a Gaussian distribution with a signal dependent variance for numerical phantoms simulating stop-and-shoot CT acquisitions. The noise model has been reported elsewhere. This experimentally verified model describes calibrated projection data, i.e. the acquired data after CT system calibration (including logarithmic transform, beam hardening, etc.). For our quantitative tests, we employed both linear Gaussian low-pass filters and non-linear L-filters with Gaussian weights. The stackgram technique provides the best resolution-noise trade-off in comparison with the radial sinogram filtering method, and thus a potential filtering approach to low-dose CT sinograms.

We discuss data filtering prior to image reconstruction. For this kind of filtering, the radial direction of the sinogram is routinely employed. Recently, we have introduced an alternative approach to sinogram data processing, exploiting the angular information in a novel way. This new stackgram representation can be regarded as an intermediate form of the sinogram and image domains. In this experimental study, we compare the radial sinogram and angular stackgram filtering methods using physical SPECT phantoms. Our study is carried out by employing simple linear and nonlinear filters with ten different Gaussian kernels, in order to provide a comparable investigation. According to our results, angular stackgram filtering with the nonlinear filters provides the best resolution-noise tradeoff of the compared methods. Besides, stackgram filtering with these filters seems to preserve the resolution in an exceptional way. Visually, noise in the reconstructed images after stackgram filtering appears more “powdery” in comparison with radial sinogram filtering.

In limited angle tomography, the projection views over a complete angular range of 180° are not available for image reconstruction. The missing part of the projection or sinogram data need to be extrapolated numerically, if standard image reconstruction methods are applied. A novel stackgram domain can be regarded as an intermediate form of the sinogram and image domains, in which the signals along the sinusoidal trajectories of a sinogram can be processed independently. In this paper, we compare extrapolation of incomplete sinogram data in the sinogram and stackgram domains along the angular directions. The extrapolated signals are assumed to be band–limited, other a priori assumptions about the data are not made. In this study, we employed simulated numerical data with different ranges of the limited projection views. According to our experiments, extrapolation of the incomplete data in the stackgram domain provides quantitatively better results as compared to extrapolation in the sinogram domain. In addition, tangential degradation in the reconstructed images can not be observed in the case of stackgram extrapolation, in contrast to angular sinogram extrapolation.

A sinogram is image representation of raw data obtained from projections of the object for image reconstruction. Radial sinogram domain filtering, i.e. filtering along the projections of the sinogram, is a common noise reduction technique for the data. The stackgram domain, in which the signals along the sinusoidal trajectories of the sinogram can be processed separately, is a new method for sinogram data filtering. We have shown in our earlier studies the potential of the stackgram domain approach, but not its superiority compared to the other filtering methods. In this study, we compare radial sinogram domain and angular stackgram domain filtering techniques employing nonlinear L-filters. According to our quantitative studies, that are also comparable to our previous studies, stackgram domain filtering provides the best resolution versus noise trade‐off for the compared methods. Besides, as noticed in our studies, the noise structure looks visually more pleasant after stackgram filtering, compared to radial filtering.

Projection data in Positron Emission Tomography (PET) are acquired as a number of photon counts from different observation angles. Positron decay is a random phenomenon that causes undesirably high variations in measured sinogram appearing as noise. Filtering of the raw data in a stackgram domain is a new technique capable of producing the reliable estimates of underlying activity. How-ever to facilitate accurate denoising procedure of the emission data, signals constituting the stackgram should be properly modelled. In this study the choice of appropriate noise model for them is considered. We will demonstrate that the general two-parameter distributions can be employed to properly characterize the source of errors coming from the data measurement process and from the decay process of the positron emitting tracer.