Receive a weekly summary and discussion of the top papers of the week by leading researchers in the field.

General General

Intensive care photoplethysmogram datasets and machine-learning for blood pressure estimation: Generalization not guarantied.

In Frontiers in physiology

The large MIMIC waveform dataset, sourced from intensive care units, has been used extensively for the development of Photoplethysmography (PPG) based blood pressure (BP) estimation algorithms. Yet, because the data comes from patients in severe conditions-often under the effect of drugs-it is regularly noted that the relationship between BP and PPG signal characteristics may be anomalous, a claim that we investigate here. A sample of 12,000 records from the MIMIC waveform dataset was stacked up against the 219 records of the PPG-BP dataset, an alternative public dataset obtained under controlled experimental conditions. The distribution of systolic and diastolic BP data and 31 PPG pulse morphological features was first compared between datasets. Then, the correlation between features and BP, as well as between the features themselves, was analysed. Finally, regression models were trained for each dataset and validated against the other. Statistical analysis showed significant p < 0.001 differences between the datasets in diastolic BP and in 20 out of 31 features when adjusting for heart rate differences. The eight features showing the highest rank correlation ρ   >   0.40 to systolic BP in PPG-BP all displayed muted correlation levels ρ   <   0.10 in MIMIC. Regression tests showed twice higher baseline predictive power with PPG-BP than with MIMIC. Cross-dataset regression displayed a practically complete loss of predictive power for all models. The differences between the MIMIC and PPG-BP dataset exposed in this study suggest that BP estimation models based on the MIMIC dataset have reduced predictive power on the general population.

Weber-Boisvert Guillaume, Gosselin Benoit, Sandberg Frida


BP estimation, PPG datasets, PPG-BP, UCI, blood pressure estimation, intensive care datasets, mimic, photoplethysmography

General General

Using machine learning to identify early predictors of adolescent emotion regulation development.

In Journal of research on adolescence : the official journal of the Society for Research on Adolescence

As 20% of adolescents develop emotion regulation difficulties, it is important to identify important early predictors thereof. Using the machine learning algorithm SEM-forests, we ranked the importance of (87) candidate variables assessed at age 13 in predicting quadratic latent trajectory models of emotion regulation development from age 14 to 18. Participants were 497 Dutch families. Results indicated that the most important predictors were individual differences (e.g., in personality), aspects of relationship quality and conflict behaviors with parents and peers, and internalizing and externalizing problems. Relatively less important were demographics, bullying, delinquency, substance use, and specific parenting practices-although negative parenting practices ranked higher than positive ones. We discuss implications for theory and interventions, and present an open source risk assessment tool, ERRATA.

Van Lissa Caspar J, Beinhauer Lukas, Branje Susan, Meeus Wim H J


adolescence, emotion regulation, machine learning, random forests, theory formation

General General

Obstacles to effective model deployment in healthcare.

In Journal of bioinformatics and computational biology

Despite an exponential increase in publications on clinical prediction models over recent years, the number of models deployed in clinical practice remains fairly limited. In this paper, we identify common obstacles that impede effective deployment of prediction models in healthcare, and investigate their underlying causes. We observe a key underlying cause behind most obstacles - the improper development and evaluation of prediction models. Inherent heterogeneities in clinical data complicate the development and evaluation of clinical prediction models. Many of these heterogeneities in clinical data are unreported because they are deemed to be irrelevant, or due to privacy concerns. We provide real-life examples where failure to handle heterogeneities in clinical data, or sources of biases, led to the development of erroneous models. The purpose of this paper is to familiarize modeling practitioners with common sources of biases and heterogeneities in clinical data, both of which have to be dealt with to ensure proper development and evaluation of clinical prediction models. Proper model development and evaluation, together with complete and thorough reporting, are important prerequisites for a prediction model to be effectively deployed in healthcare.

Chan Wei Xin, Wong Limsoon


Clinical prediction models, deployment, machine learning

General General

Asynchrony rescues statistically optimal group decisions from information cascades through emergent leaders.

In Royal Society open science

It is usually assumed that information cascades are most likely to occur when an early but incorrect opinion spreads through the group. Here, we analyse models of confidence-sharing in groups and reveal the opposite result: simple but plausible models of naive-Bayesian decision-making exhibit information cascades when group decisions are synchronous; however, when group decisions are asynchronous, the early decisions reached by Bayesian decision-makers tend to be correct and dominate the group consensus dynamics. Thus early decisions actually rescue the group from making errors, rather than contribute to it. We explore the likely realism of our assumed decision-making rule with reference to the evolution of mechanisms for aggregating social information, and known psychological and neuroscientific mechanisms.

Reina Andreagiovanni, Bose Thomas, Srivastava Vaibhav, Marshall James A R


Bayesian brain, collective decision-making, emergent leaders, information cascades

Pathology Pathology

Uncovering spatiotemporal patterns of atrophy in progressive supranuclear palsy using unsupervised machine learning.

In Brain communications

To better understand the pathological and phenotypic heterogeneity of progressive supranuclear palsy and the links between the two, we applied a novel unsupervised machine learning algorithm (Subtype and Stage Inference) to the largest MRI data set to date of people with clinically diagnosed progressive supranuclear palsy (including progressive supranuclear palsy-Richardson and variant progressive supranuclear palsy syndromes). Our cohort is comprised of 426 progressive supranuclear palsy cases, of which 367 had at least one follow-up scan, and 290 controls. Of the progressive supranuclear palsy cases, 357 were clinically diagnosed with progressive supranuclear palsy-Richardson, 52 with a progressive supranuclear palsy-cortical variant (progressive supranuclear palsy-frontal, progressive supranuclear palsy-speech/language, or progressive supranuclear palsy-corticobasal), and 17 with a progressive supranuclear palsy-subcortical variant (progressive supranuclear palsy-parkinsonism or progressive supranuclear palsy-progressive gait freezing). Subtype and Stage Inference was applied to volumetric MRI features extracted from baseline structural (T1-weighted) MRI scans and then used to subtype and stage follow-up scans. The subtypes and stages at follow-up were used to validate the longitudinal consistency of subtype and stage assignments. We further compared the clinical phenotypes of each subtype to gain insight into the relationship between progressive supranuclear palsy pathology, atrophy patterns, and clinical presentation. The data supported two subtypes, each with a distinct progression of atrophy: a 'subcortical' subtype, in which early atrophy was most prominent in the brainstem, ventral diencephalon, superior cerebellar peduncles, and the dentate nucleus, and a 'cortical' subtype, in which there was early atrophy in the frontal lobes and the insula alongside brainstem atrophy. There was a strong association between clinical diagnosis and the Subtype and Stage Inference subtype with 82% of progressive supranuclear palsy-subcortical cases and 81% of progressive supranuclear palsy-Richardson cases assigned to the subcortical subtype and 82% of progressive supranuclear palsy-cortical cases assigned to the cortical subtype. The increasing stage was associated with worsening clinical scores, whilst the 'subcortical' subtype was associated with worse clinical severity scores compared to the 'cortical subtype' (progressive supranuclear palsy rating scale and Unified Parkinson's Disease Rating Scale). Validation experiments showed that subtype assignment was longitudinally stable (95% of scans were assigned to the same subtype at follow-up) and individual staging was longitudinally consistent with 90% remaining at the same stage or progressing to a later stage at follow-up. In summary, we applied Subtype and Stage Inference to structural MRI data and empirically identified two distinct subtypes of spatiotemporal atrophy in progressive supranuclear palsy. These image-based subtypes were differentially enriched for progressive supranuclear palsy clinical syndromes and showed different clinical characteristics. Being able to accurately subtype and stage progressive supranuclear palsy patients at baseline has important implications for screening patients on entry to clinical trials, as well as tracking disease progression.

Scotton William J, Shand Cameron, Todd Emily, Bocchetta Martina, Cash David M, VandeVrede Lawren, Heuer Hilary, Young Alexandra L, Oxtoby Neil, Alexander Daniel C, Rowe James B, Morris Huw R, Boxer Adam L, Rohrer Jonathan D, Wijeratne Peter A


Subtype and Stage Inference, biomarkers, disease progression, machine learning, progressive supranuclear palsy

Ophthalmology Ophthalmology

CMS-NET: deep learning algorithm to segment and quantify the ciliary muscle in swept-source optical coherence tomography images.

In Therapeutic advances in chronic disease

BACKGROUND : The ciliary muscle plays a role in changing the shape of the crystalline lens to maintain the clear retinal image during near work. Studying the dynamic changes of the ciliary muscle during accommodation is necessary for understanding the mechanism of presbyopia. Optical coherence tomography (OCT) has been frequently used to image the ciliary muscle and its changes during accommodation in vivo. However, the segmentation process is cumbersome and time-consuming due to the large image data sets and the impact of low imaging quality.

OBJECTIVES : This study aimed to establish a fully automatic method for segmenting and quantifying the ciliary muscle on the basis of optical coherence tomography (OCT) images.

DESIGN : A perspective cross-sectional study.

METHODS : In this study, 3500 signed images were used to develop a deep learning system. A novel deep learning algorithm was created from the widely used U-net and a full-resolution residual network to realize automatic segmentation and quantification of the ciliary muscle. Finally, the algorithm-predicted results and manual annotation were compared.

RESULTS : For segmentation performed by the system, the total mean pixel value difference (PVD) was 1.12, and the Dice coefficient, intersection over union (IoU), and sensitivity values were 93.8%, 88.7%, and 93.9%, respectively. The performance of the system was comparable with that of experienced specialists. The system could also successfully segment ciliary muscle images and quantify ciliary muscle thickness changes during accommodation.

CONCLUSION : We developed an automatic segmentation framework for the ciliary muscle that can be used to analyze the morphological parameters of the ciliary muscle and its dynamic changes during accommodation.

Chen Wen, Yu Xiangle, Ye Yiru, Gao Hebei, Cao Xinyuan, Lin Guangqing, Zhang Riyan, Li Zixuan, Wang Xinmin, Zhou Yuheng, Shen Meixiao, Shao Yilei


accommodation, ciliary muscle, deep learning, optical coherence tomography (OCT), presbyopia