•Investigating the effect of image downsampling and cropping on skin lesion classification performance.•Proposing a three-level fusion approach using multiple fine-tuned deep networks and cropped ...multi-scale skin lesion images.•Achieving excellent classification performance on the ISIC 2018 challenge test dataset.
Background and objective: Skin cancer is among the most common cancer types in the white population and consequently computer aided methods for skin lesion classification based on dermoscopic images are of great interest. A promising approach for this uses transfer learning to adapt pre-trained convolutional neural networks (CNNs) for skin lesion diagnosis. Since pre-training commonly occurs with natural images of a fixed image resolution and these training images are usually significantly smaller than dermoscopic images, downsampling or cropping of skin lesion images is required. This however may result in a loss of useful medical information, while the ideal resizing or cropping factor of dermoscopic images for the fine-tuning process remains unknown.
Methods: We investigate the effect of image size for skin lesion classification based on pre-trained CNNs and transfer learning. Dermoscopic images from the International Skin Imaging Collaboration (ISIC) skin lesion classification challenge datasets are either resized to or cropped at six different sizes ranging from 224 × 224 to 450 × 450. The resulting classification performance of three well established CNNs, namely EfficientNetB0, EfficientNetB1 and SeReNeXt-50 is explored. We also propose and evaluate a multi-scale multi-CNN (MSM-CNN) fusion approach based on a three-level ensemble strategy that utilises the three network architectures trained on cropped dermoscopic images of various scales.
Results: Our results show that image cropping is a better strategy compared to image resizing delivering superior classification performance at all explored image scales. Moreover, fusing the results of all three fine-tuned networks using cropped images at all six scales in the proposed MSM-CNN approach boosts the classification performance compared to a single network or a single image scale. On the ISIC 2018 skin lesion classification challenge test set, our MSM-CNN algorithm yields a balanced multi-class accuracy of 86.2% making it the currently second ranked algorithm on the live leaderboard.
Conclusions: We confirm that the image size has an effect on skin lesion classification performance when employing transfer learning of CNNs. We also show that image cropping results in better performance compared to image resizing. Finally, a straightforward ensembling approach that fuses the results from images cropped at six scales and three fine-tuned CNNs is shown to lead to the best classification performance.
Research on machine learning approaches for upper-limb prosthesis control has shown impressive progress. However, translating these results from the lab to patient's everyday lives remains a ...challenge because advanced control schemes tend to break down under everyday disturbances, such as electrode shifts. Recently, it has been suggested to apply adaptive transfer learning to counteract electrode shifts using as little newly recorded training data as possible. In this paper, we present a novel, simple version of transfer learning and provide the first user study demonstrating the effectiveness of transfer learning to counteract electrode shifts. For this purpose, we introduce the novel Box and Beans test to evaluate prosthesis proficiency and compare user performance with an initial simple pattern recognition system, the system under electrode shifts, and the system after transfer learning. Our results show that transfer learning could significantly alleviate the impact of electrode shifts on user performance in the Box and Beans test.
Coordinated shifts of neuronal activity in the prefrontal cortex are associated with strategy adaptations in behavioural tasks, when animals switch from following one rule to another. However, ...network dynamics related to multiple-rule changes are scarcely known. We show how firing rates of individual neurons in the prelimbic and cingulate cortex correlate with the performance of rats trained to change their navigation multiple times according to allocentric and egocentric strategies. The concerted population activity exhibits a stable firing during the performance of one rule but shifted to another neuronal firing state when a new rule is learnt. Interestingly, when the same rule is presented a second time within the same session, neuronal firing does not revert back to the original neuronal firing state, but a new activity-state is formed. Our data indicate that neuronal firing of prefrontal cortical neurons represents changes in strategy and task-performance rather than specific strategies or rules.
With the advent of digital pathology and microscopic systems that can scan and save whole slide histological images automatically, there is a growing trend to use computerized methods to analyze ...acquired images. Among different histopathological image analysis tasks, nuclei instance segmentation plays a fundamental role in a wide range of clinical and research applications. While many semi- and fully-automatic computerized methods have been proposed for nuclei instance segmentation, deep learning (DL)-based approaches have been shown to deliver the best performances. However, the performance of such approaches usually degrades when tested on unseen datasets. In this work, we propose a novel method to improve the generalization capability of a DL-based automatic segmentation approach. Besides utilizing one of the state-of-the-art DL-based models as a baseline, our method incorporates non-deterministic train time and deterministic test time stain normalization, and ensembling to boost the segmentation performance. We trained the model with one single training set and evaluated its segmentation performance on seven test datasets. Our results show that the proposed method provides up to 4.9%, 5.4%, and 5.9% better average performance in segmenting nuclei based on Dice score, aggregated Jaccard index, and panoptic quality score, respectively, compared to the baseline segmentation model.
While machine learning (ML) algorithms may predict cardiovascular outcomes more accurately than statistical models, their result is usually not representable by a transparent formula. Hence, it is ...often unclear how specific values of predictors lead to the predictions. We aimed to demonstrate with graphical tools how predictor-risk relations in cardiovascular risk prediction models fitted by ML algorithms and by statistical approaches may differ, and how sample size affects the stability of the estimated relations.
We reanalyzed data from a large registry of 1.5 million participants in a national health screening program. Three data analysts developed analytical strategies to predict cardiovascular events within 1 year from health screening. This was done for the full data set and with gradually reduced sample sizes, and each data analyst followed their favorite modeling approach. Predictor-risk relations were visualized by partial dependence and individual conditional expectation plots.
When comparing the modeling algorithms, we found some similarities between these visualizations but also occasional divergence. The smaller the sample size, the more the predictor-risk relation depended on the modeling algorithm used, and also sampling variability played an increased role. Predictive performance was similar if the models were derived on the full data set, whereas smaller sample sizes favored simpler models.
Predictor-risk relations from ML models may differ from those obtained by statistical models, even with large sample sizes. Hence, predictors may assume different roles in risk prediction models. As long as sample size is sufficient, predictive accuracy is not largely affected by the choice of algorithm.
Digitalisation is changing all areas of our daily life. This changing environment requires new competences from physicians in all specialities. This study systematically surveyed the knowledge, ...attitude, and interests of medical students. These results will help further develop the medical curriculum, as well as increase our understanding of future physicians by other healthcare market players. A web-based survey consisting of four sections was developed: Section one queried demographic data, section two assessed the current digital health knowledge of medical students, section three queried their attitudes about the future impact of digital health in medicine and section four assessed the recommendations medical students have for the medical curriculum in terms of digital health. This survey was distributed to all (11,978) student at all public Austrian medical schools. A total of 8.4% of the medical student population started the survey. At the knowledge self-assessment section, the medical students reached mean of 11.74 points (SD 4.42) out of a possible maximum of 32 (female mean 10.66/ SD 3.87, male mean 13.34/SD 4.50). The attitude section showed that students see digitalisation as a threat, especially with respect to the patient-physician relationship. The curriculum recommendation section showed a high interest for topics related to AI, a per study year increasing interest in impact of digital health in communication, as well as a decreasing interest in robotic related topics. The attitude towards digital health can be described as sceptical. To ensure that future physicians keep pace with this development and fulfil their responsibility towards the society, medical schools need to be more proactive to foster the understanding of medical students that digital health will persistently alter the medical practice.
Neuronal signals in the prefrontal cortex have been reported to predict upcoming decisions. Such activity patterns are often coupled to perceptual cues indicating correct choices or values of ...different options. How does the prefrontal cortex signal future decisions when no cues are present but when decisions are made based on internal valuations of past experiences with stochastic outcomes? We trained rats to perform a two-arm bandit-task, successfully adjusting choices between certain-small or possible-big rewards with changing long-term advantages. We discovered specialized prefrontal neurons, whose firing during the encounter of no-reward predicted the subsequent choice of animals, even for unlikely or uncertain decisions and several seconds before choice execution. Optogenetic silencing of the prelimbic cortex exclusively timed to encounters of no reward, provoked animals to excessive gambling for large rewards. Firing of prefrontal neurons during outcome evaluation signals subsequent choices during gambling and is essential for dynamically adjusting decisions based on internal valuations.
•Activity of prelimbic neurons predict upcoming decisions during gambling•Future choice –predictive firing occurs during evaluation of current outcome•Time-specific inactivation of prelimbic cortex increases high-risk gambling behavior•Prelimbic neurons contribute to adjusting decisions based on internal valuations
Passecker et al. show that specialized neurons in the prelimbic cortex of rats predict the next choice during the outcome evaluation in a gambling task, even for unlikely or uncertain decisions. Disrupting the prelimbic cortex led to excessive risk taking.
To investigate differences between visual sleep scoring according to the classification developed by Rechtschaffen and Kales (R&K, 1968) and scoring based on the new guidelines of the American ...Academy of Sleep Medicine (AASM, 2007).
All-night polysomnographic recordings were scored visually according to the R&K and AASM rules by experienced sleep scorers. Descriptive data analysis was used to compare the resulting sleep parameters.
Healthy subjects and patients (38 females and 34 males) aged between 21 and 86 years.
N/A.
While sleep latency and REM latency, total sleep time, and sleep efficiency were not affected by the classification standard, the time (in minutes and in percent of total sleep time) spent in sleep stage 1 (S1/N1), stage 2 (S2/N2) and slow wave sleep (S3+S4/N3) differed significantly between the R&K and the AASM classification. While light and deep sleep increased (S1 vs. N1 +10.6 min, (+2.8%): P<0.01; S3+S4 vs. N3 +9.1 min (+2.4%): P<0.01), stage 2 sleep decreased significantly according to AASM rules (S2 vs. N2 -20.5 min, (-4.9%): P<0.01). Moreover, wake after sleep onset was significantly prolonged by approximately 4 minutes (P<0.01) according to the AASM standard. Interestingly, the effects on stage REM were age-dependent (intercept at 20 years: -7.5 min; slope: 1.6 min for 10-year age increase). No effects of sex and diagnosis were observed.
The study shows significant and age-dependent differences between sleep parameters derived from conventional visual sleep scorings on the basis of R&K rules and those based on the new AASM rules. Thus, new normative data have to be established for the AASM standard.
•We search for objective sleep quality measures.•We model sleep as a continuous process represented by a set of sleep microstates.•We correlate parameters of the sleep model with latent factors of ...daytime behavioral and life-quality measures.•We report a set of sleep parameters significantly correlating with daytime factors.
The main goal of this study was to investigate to what extent polysomnographic (PSG) recordings of nocturnal human sleep can provide information about sleep quality in terms of correlation with a set of daytime measures. These measures were designed with the aim of comprising selected quality of night sleep and consist of subjective sleep quality ratings, neuropsychological tests and physiological parameters. First, a factor analysis model was applied to the large number of daytime measures of sleep quality in order to detect their latent structure. Secondly, in addition to the gold standard sleep staging method to arrive at variables about sleep architecture from PSG, we applied a recently developed continuous sleep representation by considering the probabilistic sleep model (PSM) describing the microstructure of sleep. Significant correlations between sleep architecture and daytime variables of sleep quality were found. Both the factor analysis and the PSM helped maximize the information about this relationship.
Summary
Interrater variability of sleep stage scorings has an essential impact not only on the reading of polysomnographic sleep studies (PSGs) for clinical trials but also on the evaluation of ...patients’ sleep. With the introduction of a new standard for sleep stage scorings (AASM standard) there is a need for studies on interrater reliability (IRR). The SIESTA database resulting from an EU‐funded project provides a large number of studies (n = 72; 56 healthy controls and 16 subjects with different sleep disorders, mean age ± SD: 57.7 ± 18.7, 34 females) for which scorings according to both standards (AASM and R&K) were done. Differences in IRR were analysed at two levels: (1) based on quantitative sleep parameter by means of intraclass correlations; and (2) based on an epoch‐by‐epoch comparison by means of Cohen’s kappa and Fleiss’ kappa. The overall agreement was for the AASM standard 82.0% (Cohen’s kappa = 0.76) and for the R&K standard 80.6% (Cohen’s kappa = 0.68). Agreements increased from R&K to AASM for all sleep stages, except N2. The results of this study underline that the modification of the scoring rules improve IRR as a result of the integration of occipital, central and frontal leads on the one hand, but decline IRR on the other hand specifically for N2, due to the new rule that cortical arousals with or without concurrent increase in submental electromyogram are critical events for the end of N2.