An Optimized Approach for Prostate Image Segmentation Using K-Means Clustering Algorithm with Elbow Method

Out-of-distribution generalization for segmentation of lymph node metastasis in breast cancer

Pathology provides the definitive diagnosis, and Artificial Intelligence (AI) tools are poised to improve accuracy, inter-rater agreement, and turn-around time (TAT) of pathologists, leading to improved quality of care. A high value clinical application is the grading of Lymph Node Metastasis (LNM) which is used for breast cancer staging and guides treatment decisions. A challenge of implementing AI tools widely for LNM classification is domain shift, where Out-of-Distribution (OOD) data has a different distribution than the In-Distribution (ID) data used to train the model, resulting in a drop in performance in OOD data. This work proposes a novel clustering and sampling method to automatically curate training datasets in an unsupervised manner with the aim of improving model generalization abilities. To evaluate the generalization performance of the proposed models, we applied a novel use of the Two One-sided Tests (TOST) method. This method examines whether the performance on ID and OOD data is equivalent, serving as a proxy for generalization. We provide the first evidence for computing equivalence margins that are data-dependent, which reduces subjectivity. The proposed framework shows the ensembled models constructed from models that generalized across both tumor and normal patches enhanced performance, achieving an F1 score of 0.81 for LNM classification on unseen ID and OOD samples. Interactive viewing of slide-level segmentations can be accessed on PathcoreFlow™ through https://web.pathcore.com/folder/18555?s=QTJVHJuhrfe5 . Segmentation models are available at https://github.com/IAMLAB-Ryerson/OOD-Generalization-LNM .

Identification of severe acute pediatric asthma phenotypes using unsupervised machine learning

RATIONALE

More targeted management of severe acute pediatric asthma could improve clinical outcomes.

OBJECTIVES

To identify distinct clinical phenotypes of severe acute pediatric asthma using variables obtained in the first 12 h of hospitalization.

METHODS

We conducted a retrospective cohort study in a quaternary care children's hospital from 2014 to 2022. Encounters for children ages 2-18 years admitted to the hospital for asthma were included. We used consensus k means clustering with patient demographics, vital signs, diagnostics, and laboratory data obtained in the first 12 h of hospitalization.

MEASUREMENTS AND MAIN RESULTS

The study population included 683 encounters divided into derivation (80%) and validation (20%) sets, and two distinct clusters were identified. Compared to Cluster 1 in the derivation set, Cluster 2 encounters (177 [32%]) were older (11 years [8; 14] vs. 5 years [3; 8]; p < .01) and more commonly males (63% vs. 53%; p = .03) of Black race (51% vs. 40%; p = .03) with non-Hispanic ethnicity (96% vs. 84%; p < .01). Cluster 2 encounters had smaller improvements in vital signs at 12-h including percent change in heart rate (-1.7 [-11.7; 12.7] vs. -7.8 [-18.5; 1.7]; p < .01), and respiratory rate (0.0 [-20.0; 22.2] vs. -11.4 [-27.3; 9.0]; p < .01). Encounters in Cluster 2 had lower percentages of neutrophils (70.0 [55.0; 83.0] vs. 85.0 [77.0; 90.0]; p < .01) and higher percentages of lymphocytes (17.0 [8.0; 32.0] vs. 9.0 [5.3; 14.0]; p < .01). Cluster 2 encounters had higher rates of invasive mechanical ventilation (23% vs. 5%; p < .01), longer hospital length of stay (4.5 [2.6; 8.8] vs. 2.9 [2.0; 4.3]; p < .01), and a higher mortality rate (7.3% vs. 0.0%; p < .01). The predicted cluster assignments in the validation set shared the same ratio (~2:1), and many of the same characteristics.

CONCLUSIONS

We identified two clinical phenotypes of severe acute pediatric asthma which exhibited distinct clinical features and outcomes.

Modeling and prediction of set‑up errors in breast cancer image‑guided radiotherapy using the Gaussian mixture model

The aim of the present study was to develop a prediction model for set-up error distribution in breast cancer image-guided radiotherapy (IGRT) using a Gaussian mixture model (GMM). To achieve this, the image-guided set-up errors data of 80 patients with breast cancer were selected, and the GMM was used to develop the set-up errors distribution prediction model. The predicted error center points, covariance and probability were calculated and compared with the planning target volume (PTV) margin formula. A total of 1,200 sets of set-up errors in IGRT for breast cancer were collected. The results of the Gaussian model parameters showed that the set-up errors were mainly in the direction of µ1-µ4 center points. All the raw errors in the lateral, longitudinal and vertical directions were -6.30-4.60, -5.40-1.47 and -2.70-1.70 mm, respectively. According to the probability of each center, the set-up error was most likely to shift in the µ1 direction, reaching 0.53. The set-up errors of the other three centers, µ2, µ3 and µ4, were 0.11, 0.34 and 0.12, respectively. According to the covariance parameters of the GMM, the maximum statistical standard deviation of the set-up errors reached 29.06. In conclusion, the results of the present study demonstrated that the GMM can be used to quantitatively describe and predict the distribution of set-up errors in IGRT for breast cancer, and these findings could be useful as a reference for set-up error control and tumor PTV expansion in breast cancer radiotherapy without routine, daily IGRT.

Decoding Radiomics: A Step-by-Step Guide to Machine Learning Workflow in Hand-Crafted and Deep Learning Radiomics Studies

Although radiomics research has experienced rapid growth in recent years, with numerous studies dedicated to the automated extraction of diagnostic and prognostic information from various imaging modalities, such as CT, PET, and MRI, only a small fraction of these findings has successfully transitioned into clinical practice. This gap is primarily due to the significant methodological challenges involved in radiomics research, which emphasize the need for a rigorous evaluation of study quality. While many technical aspects may lie outside the expertise of most radiologists, having a foundational knowledge is essential for evaluating the quality of radiomics workflows and contributing, together with data scientists, to the development of models with a real-world clinical impact. This review is designed for the new generation of radiologists, who may not have specialized training in machine learning or radiomics, but will inevitably play a role in this evolving field. The paper has two primary objectives: first, to provide a clear, systematic guide to radiomics study pipeline, including study design, image preprocessing, feature selection, model training and validation, and performance evaluation. Furthermore, given the critical importance of evaluating the robustness of radiomics studies, this review offers a step-by-step guide to the application of the METhodological RadiomICs Score (METRICS, 2024)-a newly proposed tool for assessing the quality of radiomics studies. This roadmap aims to support researchers and reviewers alike, regardless of their machine learning expertise, in utilizing this tool for effective study evaluation.

Cluster analysis of adult individuals with type 1 diabetes: Treatment pathways and complications over a five-year follow-up period

AIMS

To identify subgroups of adults with type 1 diabetes and analyse their treatment pathways and risk of diabetes-related complications over a 5-year follow-up.

METHODS

We performed a k-means cluster analysis using the T1DExchange Registry (n = 6,302) to identify subgroups based on demographic and clinical characteristics. Annual reassessments linked treatment trajectories with these clusters, considering drug and technology use. Complication risks were analysed using Cox regression.

RESULTS

Five clusters were identified: 1) A favourable combination of all variables (31.67 %); 2) Longer diabetes duration (22.63 %); 3) Higher HbA1c levels (13.28 %); 4) Higher BMI (15.25 %); 5) Older age at diagnosis (17.17 %). Two-thirds of patients remained in their initial cluster annually. Technology adoption showed improved glycaemic control over time. Cox proportional hazards showed different risk patterns: Cluster 1 had low complication risk; Cluster 2 had the highest risk for retinopathy, coronary artery disease and autonomic neuropathy; Cluster 3 had the highest risk for albuminuria, depression and diabetic ketoacidosis; Cluster 4 had increased risk for multiple complications; Cluster 5 had the highest risk for hypertension and severe hypoglycaemia, with elevated coronary artery disease risk.

CONCLUSIONS

Clinical characteristics can identify subgroups of patients with T1DM showing differences in treatment and complications during follow-up.

Automatic Detection of Scalp High-Frequency Oscillations Based on Deep Learning

Scalp high-frequency oscillations (sHFOs) are a promising non-invasive biomarker of epilepsy. However, the visual marking of sHFOs is a time-consuming and subjective process, existing automatic detectors based on single-dimensional analysis have difficulty with accurately eliminating artifacts and thus do not provide sufficient reliability to meet clinical needs. Therefore, we propose a high-performance sHFOs detector based on a deep learning algorithm. An initial detection module was designed to extract candidate high-frequency oscillations. Then, one-dimensional (1D) and two-dimensional (2D) deep learning models were designed, respectively. Finally, the weighted voting method is used to combine the outputs of the two model. In experiments, the precision, recall, specificity and F1-score were 83.44%, 83.60%, 96.61% and 83.42%, respectively, on average and the kappa coefficient was 80.02%. In addition, the proposed detector showed a stable performance on multi-centre datasets. Our sHFOs detector demonstrated high robustness and generalisation ability, which indicates its potential applicability as a clinical assistance tool. The proposed sHFOs detector achieves an accurate and robust method via deep learning algorithm.

Association of childhood adversities with psychosocial difficulties among Chinese children and adolescents

BACKGROUND

Adverse childhood experiences (ACEs) have been well recognized as risk factors for various adverse outcomes. However, the impacts of ACEs on psychological wellbeing among Chinese children and adolescents are unknown.

METHODS

In total, 27 414 participants (6592 Grade 4-6 and 20 822 Grade 7-12 students) were included and information on ACEs and various psychosocial outcomes was collected. We identified subgroups with distinct psychosocial statuses using cluster analysis and logistic regression was applied to measure the associations of ACEs [individual, cumulative numbers by categories or co-occurring patterns identified by using multiple correspondence analysis (MCA)] with item- and cluster-specific psychosocial difficulties.

RESULTS

Three and four cluster-based psychosocial statuses were identified for Grade 4-6 and Grade 7-12 students, respectively, indicating that psychosocial difficulties among younger students were mainly presented as changes in relationships/behaviours, whereas older students were more likely featured by deviations in multiple domains including psychiatric symptoms and suicidality. Strongest associations were found for threat-related ACEs (e.g. bullying experiences) with item- or cluster-based psychosocial difficulties (e.g. for cluster-based difficulties, the highest odds ratios = 1.72-2.08 for verbal bullying in Grade 4-6 students and 6.30-12.81 for cyberbullying in Grade 7-12 students). Analyses on cumulative numbers of ACEs and MCA-based ACE patterns revealed similar risk patterns. Additionally, exposure patterns predominated by poor external environment showed significant associations with psychosocial difficulties among Grade 7-12 students but not Grade 4-6 students.

CONCLUSIONS

Chinese adolescents faced different psychosocial difficulties that varied by age, all of which were associated with ACEs, particularly threat-related ACEs. Such findings prompt the development of early interventions for those key ACEs to prevent psychosocial adversities among children and adolescents.

A novel planning framework for efficient spot-scanning proton arc therapy via particle swarm optimization (SPArc-particle swarm)

Objective.Delivery efficiency is the bottleneck of spot-scanning proton arc therapy (SPArc) because of the numerous energy layers (ELs) ascending switches. This study aims to develop a new algorithm to mitigate the need for EL ascending via water equivalent thickness (WET) sector selection followed by particle swarm optimization (SPArc-particle swarm).Approach.SPArc-particle swarmdivided the full arc trajectory into the optimal sectors based on K-means clustering analysis of the relative mean WET. Within the sector, particle swarm optimization was used to minimize the total energy switch time, optimizing the energy selection integrated with the EL delivery sequence and relationship. This novel planning framework was implemented on the open-source platform matRad (Department of Medical Physics in Radiation Oncology, German Cancer Research Center-DKFZ). Three representative cases (brain, liver, and prostate cancer) were selected for testing purposes. Two kinds of plans were generated: SPArc_seq and SPArc-particle swarm. The plan quality and delivery efficiency were evaluated.Main results. With a similar plan quality, the delivery efficiency was significantly improved using SPArc-particle swarmcompared to SPArc_seq. More specifically, it reduces the number of ELs ascending switching compared to the SPArc_seq (from 21 to 7 in the brain, from 21 to 5 in the prostate, from 21 to 6 in the liver), leading to a 16%-26% reduction of the beam delivery time (BDT) in the SPArc treatment.Significance. A novel planning framework, SPArc-particle swarm, could significantly improve the delivery efficiency, which paves the roadmap towards routine clinical implementation.

A Clinical Decision Tool for Risk Stratifying Patients with Systemic Sclerosis-Related Pulmonary Hypertension

We devised a scoring system to identify patients with systemic sclerosis (SSc) at risk for pulmonary hypertension (PH) and predict all-cause mortality. Using 7 variables obtained via pulmonary function testing, echocardiography, and computed tomographic chest imaging, we applied the score to a retrospective cohort of 117 patients with SSc. There were 60 (51.3%) who were diagnosed with PH by right heart catheterization. Using a scoring threshold ≥ 0, our decision tool predicted PH with a sensitivity, specificity, and accuracy of 0.87 (95% CI 0.75, 0.94), 0.74 (95% CI 0.60, 0.84), and 0.80 (95% CI 0.72, 0.87), respectively. When adjusted for age at PH diagnosis, sex, and receipt of pulmonary arterial vasodilators, each one-point score increase was associated with an adjusted HR of 1.19 (95% CI 1.05, 1.34) for all-cause mortality. With further validation in external cohorts, our simplified clinical decision tool may better streamline earlier detection of PH in SSc.

A classification of the plantar intrinsic foot muscles based on the physiological cross-sectional area and muscle fiber length in healthy young adult males

BACKGROUND

Plantar intrinsic foot muscles (PIFMs) are composed of 10 muscles and play an essential role in achieving functional diversity in the foot. Previous studies have identified that the morphological profiles of PIFMs vary between individuals. The morphological profiles of a muscle theoretically reflect its output potentials: the physiological cross-sectional area (PCSA) of a muscle is proportional to its maximum force generation, and the muscle fiber length (FL) is its shortening velocity. This implies that the PCSA and FL may be useful variables for characterizing the functional diversity of the individual PIFM. The purpose of this study was to examine how individual PIFMs can be classified based on their PCSA and FL.

METHODS

In 26 healthy young adult males, the muscle volume and muscle length of seven PIFMs (abductor hallucis, ABDH; abductor digiti minimi, ABDM; adductor hallucis oblique head, ADDH-OH; ADDH transverse head, ADDH-TH; flexor digitorum brevis, FDB; flexor hallucis brevis, FHB; quadratus plantae, QP) were measured using magnetic resonance imaging. The PCSA and FL of each of the seven PIFMs were then estimated by combining the data measured from the participants and those of muscle architectural parameters documented from cadavers in previous studies. A total of 182 data samples (26 participants × 7 muscles) were classified into clusters using k-means cluster analysis. The optimal number of clusters was evaluated using the elbow method.

RESULTS

The data samples of PIFMs were assigned to four clusters with different morphological profiles: ADDH-OH and FHB, characterised by large PCSA and short FL (high force generation and slow shortening velocity potentials); ABDM and FDB, moderate PCSA and moderate FL (moderate force generation and moderate shortening velocity potentials); QP, moderate PCSA and long FL (moderate force generation and rapid shortening velocity potentials); ADDH-TH, small PCSA and moderate FL (low force generation and moderate shortening velocity potentials). ABDH components were assigned equivalently to the first and second clusters.

CONCLUSIONS

The approach adopted in this study may provide a novel perspective for interpreting the PIFMs' function based on their maximal force generation and shortening velocity potentials.

A topographical analysis of encephalocele locations: generation of a standardised atlas and cluster analysis

OBJECTIVE

Encephaloceles are considered to result from defects in the developing skull through which meninges, and potentially brain tissue, herniate. The pathological mechanism underlying this process is incompletely understood. We aimed to describe the location of encephaloceles through the generation of a group atlas to determine whether they occur at random sites or clusters within distinct anatomical regions.

METHODS

Patients diagnosed with cranial encephaloceles or meningoceles were identified from a prospectively maintained database between 1984 and 2021. Images were transformed to atlas space using non-linear registration. The bone defect, encephalocele and herniated brain contents were manually segmented allowing for a 3-dimensional heat map of encephalocele locations to be generated. The centroids of the bone defects were clustered utilising a K-mean clustering machine learning algorithm in which the elbow method was used to identify the optimal number of clusters.

RESULTS

Of the 124 patients identified, 55 had volumetric imaging in the form of MRI (48/55) or CT (7/55) that could be used for atlas generation. Median encephalocele volume was 14,704 (IQR 3655-86,746) mm3 and the median surface area of the skull defect was 679 (IQR 374-765) mm2. Brain herniation into the encephalocele was found in 45% (25/55) with a median volume of 7433 (IQR 3123-14,237) mm3. Application of the elbow method revealed 3 discrete clusters: (1) anterior skull base (22%; 12/55), (2) parieto-occipital junction (45%; 25/55) and (3) peri-torcular (33%; 18/55). Cluster analysis revealed no correlation between the location of the encephalocele with gender (χ2 (2, n = 91) = 3.86, p = 0.15). Compared to expected population frequencies, encephaloceles were relatively more common in Black, Asian and Other compared to White ethnicities. A falcine sinus was identified in 51% (28/55) of cases. Falcine sinuses were more common (χ2 (2, n = 55) = 6.09, p = 0.05) whilst brain herniation was less common (χ2 (2, n = 55) = .16.24, p < 0.0003) in the parieto-occipital location.

CONCLUSION

This analysis revealed three predominant clusters for the location of encephaloceles, with the parieto-occipital junction being the most common. The stereotypic location of encephaloceles into anatomically distinct clusters and the coexistence of distinct venous malformations at certain sites suggests that their location is not random and raises the possibility of distinct pathogenic mechanisms unique to each of these regions.

Factors associated with circulatory death after out-of-hospital cardiac arrest: a population-based cluster analysis

BACKGROUND

Out-of-hospital cardiac arrest (OHCA) is a common cause of death. Early circulatory failure is the most common reason for death within the first 48 h. This study in intensive care unit (ICU) patients with OHCA was designed to identify and characterize clusters based on clinical features and to determine the frequency of death from refractory postresuscitation shock (RPRS) in each cluster.

METHODS

We retrospectively identified adults admitted alive to ICUs after OHCA in 2011-2018 and recorded in a prospective registry for the Paris region (France). We identified patient clusters by performing an unsupervised hierarchical cluster analysis (without mode of death among the variables) based on Utstein clinical and laboratory variables. For each cluster, we estimated the hazard ratio (HRs) for RPRS.

RESULTS

Of the 4445 included patients, 1468 (33%) were discharged alive from the ICU and 2977 (67%) died in the ICU. We identified four clusters: initial shockable rhythm with short low-flow time (cluster 1), initial non-shockable rhythm with usual absence of ST-segment elevation (cluster 2), initial non-shockable rhythm with long no-flow time (cluster 3), and long low-flow time with high epinephrine dose (cluster 4). RPRS was significantly associated with this last cluster (HR, 5.51; 95% confidence interval 4.51-6.74).

CONCLUSIONS

We identified patient clusters based on Utstein criteria, and one cluster was strongly associated with RPRS. This result may help to make decisions about using specific treatments after OHCA.

Unsupervised machine learning effectively clusters pediatric spastic cerebral palsy patients for determination of optimal responders to selective dorsal rhizotomy

Selective dorsal rhizotomy (SDR) can reduce the spasticity in patients with spastic cerebral palsy (SCP) and thus improve the motor function in these patients, but different levels of improvement in motor function were observed among patients after SDR. The aim of the present study was to subgroup patients and to predict the possible outcome of SDR based on the pre-operational parameters. A hundred and thirty-five pediatric patients diagnosed with SCP who underwent SDR from January 2015 to January 2021 were retrospectively reviewed. Spasticity of lower limbs, the number of target muscles, motor functions, and other clinical parameters were used as input variables for unsupervised machine learning to cluster all included patients. The postoperative motor function change is used to assess the clinical significance of clustering. After the SDR procedure, the spasticity of muscles in all patients was reduced significantly, and the motor function was promoted significantly at the follow-up duration. All patients were categorized into three subgroups by both hierarchical and K-means clustering methods. The three subgroups showed significantly different clinical characteristics except for the age at surgery, and the post-operational motor function change at the last follow-up in these three clusters was different. Three subgroups clustered by two methods could be identified as "best responders", "good responders" and "moderate responders" based on the increasement of motor function after SDR. Clustering results achieved by hierarchical and K-means algorithms showed high consistency in subgrouping the whole group of patients. These results indicated that SDR could relieve the spasticity and promote the motor function of patients with SCP. Unsupervised machine learning methods can effectively and accurately cluster patients into different subgroups suffering from SCP based on pre-operative characteristics. Machine learning can be used for the determination of optimal responders for SDR surgery.

The global region-specific epidemiologic characteristics of influenza: World Health Organization FluNet data from 1996 to 2021

OBJECTIVES

This study aimed to investigate region-specific epidemiologic characteristics of influenza and influenza transmission zones (ITZs).

METHODS

Weekly influenza surveillance data of 156 countries from 1996 to 2021 were obtained using FluNet. Joinpoint regression was used to describe global influenza virus trends, and clustering analyses were used to classify the ITZs.

RESULTS

The global median average positive rate for total influenza virus was 16.19% (interquartile range: 11.62-25.70%). Overall, three major subtypes (influenza H1, H3, and B viruses) showed alternating epidemics. Notably, the proportion of influenza B viruses increased significantly from July 2020 to June 2021, reaching 62.66%. The primary peaks of influenza virus circulation in the north were earlier than those in the south. Global influenza virus circulation was significantly characterized by seven ITZs, including "Northern America" (primary peak: week 10), "Eastern & Southern-Asia" (primary peak: week 10), "Europe" (primary peak: week 11), "Asia-Europe" (primary peak: week 12), "Southern-America" (primary peak: week 30), "Oceania-Melanesia-Polynesia" (primary peak: week 39), and "Africa" (primary peak: week 46).

CONCLUSION

Global influenza virus circulation was significantly characterized by seven ITZs that could be applied to influenza surveillance and warning.

Fractal dimension based geographical clustering of COVID-19 time series data

Understanding the local dynamics of COVID-19 transmission calls for an approach that characterizes the incidence curve in a small geographical unit. Given that incidence curves exhibit considerable day-to-day variation, the fractal structure of the time series dynamics is investigated for the Flanders and Brussels Regions of Belgium. For each statistical sector, the smallest administrative geographical entity in Belgium, fractal dimensions of COVID-19 incidence rates, based on rolling time spans of 7, 14, and 21 days were estimated using four different estimators: box-count, Hall-Wood, variogram, and madogram. We found varying patterns of fractal dimensions across time and location. The fractal dimension is further summarized by its mean, variance, and autocorrelation over time. These summary statistics are then used to cluster regions with different incidence rate patterns using k-means clustering. Fractal dimension analysis of COVID-19 incidence thus offers important insight into the past, current, and arguably future evolution of an infectious disease outbreak.

Effect of residential relocation on environmental exposures in European cohorts: An exposome-wide approach

Residential relocation is increasingly used as a natural experiment in epidemiological studies to assess the health impact of changes in environmental exposures. Since the likelihood of relocation can be influenced by individual characteristics that also influence health, studies may be biased if the predictors of relocation are not appropriately accounted for. Using data from Swedish and Dutch adults (SDPP, AMIGO), and birth cohorts (BAMSE, PIAMA), we investigated factors associated with relocation and changes in multiple environmental exposures across life stages. We used logistic regression to identify baseline predictors of moving, including sociodemographic and household characteristics, health behaviors and health. We identified exposure clusters reflecting three domains of the urban exposome (air pollution, grey surface, and socioeconomic deprivation) and conducted multinomial logistic regression to identify predictors of exposome trajectories among movers. On average, 7 % of the participants relocated each year. Before relocating, movers were consistently exposed to higher levels of air pollution than non-movers. Predictors of moving differed between the adult and birth cohorts, highlighting the importance of life stages. In the adult cohorts, moving was associated with younger age, smoking, and lower education and was independent of cardio-respiratory health indicators (hypertension, BMI, asthma, COPD). Contrary to adult cohorts, higher parental education and household socioeconomic position were associated with a higher probability of relocation in birth cohorts, alongside being the first child and living in a multi-unit dwelling. Among movers in all cohorts, those with a higher socioeconomic position at baseline were more likely to move towards healthier levels of the urban exposome. We provide new insights into predictors of relocation and subsequent changes in multiple aspects of the urban exposome in four cohorts covering different life stages in Sweden and the Netherlands. These results inform strategies to limit bias due to residential self-selection in epidemiological studies using relocation as a natural experiment.

Data-Driven Phenotyping of Alzheimer's Disease under Epigenetic Conditions Using Partial Volume Correction of PET Studies and Manifold Learning

Alzheimer's disease (AD) is the most common form of dementia. An increasing number of studies have confirmed epigenetic changes in AD. Consequently, a robust phenotyping mechanism must take into consideration the environmental effects on the patient in the generation of phenotypes. Positron Emission Tomography (PET) is employed for the quantification of pathological amyloid deposition in brain tissues. The objective is to develop a new methodology for the hyperparametric analysis of changes in cognitive scores and PET features to test for there being multiple AD phenotypes. We used a computational method to identify phenotypes in a retrospective cohort study (532 subjects), using PET and Magnetic Resonance Imaging (MRI) images and neuropsychological assessments, to develop a novel computational phenotyping method that uses Partial Volume Correction (PVC) and subsets of neuropsychological assessments in a non-biased fashion. Our pipeline is based on a Regional Spread Function (RSF) method for PVC and a t-distributed Stochastic Neighbor Embedding (t-SNE) manifold. The results presented demonstrate that (1) the approach to data-driven phenotyping is valid, (2) the different techniques involved in the pipelines produce different results, and (3) they permit us to identify the best phenotyping pipeline. The method identifies three phenotypes and permits us to analyze them under epigenetic conditions.

Clustering of critically ill patients using an individualized learning approach enables dose optimization of mobilization in the ICU

BACKGROUND

While early mobilization is commonly implemented in intensive care unit treatment guidelines to improve functional outcome, the characterization of the optimal individual dosage (frequency, level or duration) remains unclear. The aim of this study was to demonstrate that artificial intelligence-based clustering of a large ICU cohort can provide individualized mobilization recommendations that have a positive impact on the likelihood of being discharged home.

METHODS

This study is an analysis of a prospective observational database of two interdisciplinary intensive care units in Munich, Germany. Dosage of mobilization is determined by sessions per day, mean duration, early mobilization as well as average and maximum level achieved. A k-means cluster analysis was conducted including collected parameters at ICU admission to generate clinically definable clusters.

RESULTS

Between April 2017 and May 2019, 948 patients were included. Four different clusters were identified, comprising "Young Trauma," "Severely ill & Frail," "Old non-frail" and "Middle-aged" patients. Early mobilization (< 72 h) was the most important factor to be discharged home in "Young Trauma" patients (OR_adj 10.0 [2.8 to 44.0], p < 0.001). In the cluster of "Middle-aged" patients, the likelihood to be discharged home increased with each mobilization level, to a maximum 24-fold increased likelihood for ambulating (OR_adj 24.0 [7.4 to 86.1], p < 0.001). The likelihood increased significantly when standing or ambulating was achieved in the older, non-frail cluster (OR_adj 4.7 [1.2 to 23.2], p = 0.035 and OR_adj 8.1 [1.8 to 45.8], p = 0.010).

CONCLUSIONS

An artificial intelligence-based learning approach was able to divide a heterogeneous critical care cohort into four clusters, which differed significantly in their clinical characteristics and in their mobilization parameters. Depending on the cluster, different mobilization strategies supported the likelihood of being discharged home enabling an individualized and resource-optimized mobilization approach.

TRIAL REGISTRATION

Clinical Trials NCT03666286, retrospectively registered 04 September 2018.

Effect of phosphorylation on the structural dynamics, thermal stability of human dopamine transporter: A simulation study using normal modes, molecular dynamics and Markov State Model

The Human Dopamine Transporter (hDAT) plays an essential role in modulating the Influx/Efflux of dopamine, and it is involved in the mechanism of certain neurodegenerative diseases such as Parkinson's disease. Several studies have reported important states for Dopamine transport: outward-facing open state (OFo), the outward-facing closed state (OFc), the holo-occluded state closed (holo), and the inward-facing open state (IFo). Furthermore, experimental assays have shown that different phosphorylation conditions in hDAT can affect the rate of dopamine absorption. We present a protocol using hybrid simulation methods to study the conformational dynamics and stability of states of hDAT under different phosphorylation sites. With this protocol, we explored the conformational space of hDAT, identified the states, and evaluated the free energy differences and the transition probabilities between them in each of the phosphorylation cases. We also presented the conformational changes and correlated them with those described in the literature. There is a thesis/hypothesis that the phosphorylation condition corresponding to NP-333 system (where all sites Ser/Thr from residue 2 to 62 and 254 to 613 are phosphorylated, except residue 333) would decrease the rate of dopamine transport from the extracellular medium to the intracellular medium by hDAT as previously described in the literature by Lin et al., 2003. Our results corroborated this thesis/hypothesis and the data reported. It is probably due to the affectation/changes/alteration of the conformational dynamics of this system that makes the intermediate states more likely and makes it difficult to initial states associated with the uptake of dopamine in the extracellular medium, corroborating the experimental results. Furthermore, our results showed that just single phosphorylation/dephosphorylation could alter intrinsic protein motions affecting the sampling of one or more states necessary for dopamine transport. In this sense, the modification of phosphorylation influences protein movements and conformational preferences, affecting the stability of states and the transition between them and, therefore, the transport.

Visualising spatial heterogeneity in glioblastoma using imaging habitats

Glioblastoma is a high-grade aggressive neoplasm characterised by significant intra-tumoral spatial heterogeneity. Personalising therapy for this tumour requires non-invasive tools to visualise its heterogeneity to monitor treatment response on a regional level. To date, efforts to characterise glioblastoma's imaging features and heterogeneity have focussed on individual imaging biomarkers, or high-throughput radiomic approaches that consider a vast number of imaging variables across the tumour as a whole. Habitat imaging is a novel approach to cancer imaging that identifies tumour regions or 'habitats' based on shared imaging characteristics, usually defined using multiple imaging biomarkers. Habitat imaging reflects the evolution of imaging biomarkers and offers spatially preserved assessment of tumour physiological processes such perfusion and cellularity. This allows for regional assessment of treatment response to facilitate personalised therapy. In this review, we explore different methodologies to derive imaging habitats in glioblastoma, strategies to overcome its technical challenges, contrast experiences to other cancers, and describe potential clinical applications.

Deep Learning Algorithms in the Automatic Segmentation of Liver Lesions in Ultrasound Investigations

BACKGROUND

The ultrasound is one of the most used medical imaging investigations worldwide. It is non-invasive and effective in assessing liver tumors or other types of parenchymal changes.

METHODS

The aim of the study was to build a deep learning model for image segmentation in ultrasound video investigations. The dataset used in the study was provided by the University of Medicine and Pharmacy Craiova, Romania and contained 50 video examinations from 49 patients. The mean age of the patients in the cohort was 69.57. Regarding presence of a subjacent liver disease, 36.73% had liver cirrhosis and 16.32% had chronic viral hepatitis (5 patients: chronic hepatitis C and 3 patients: chronic hepatitis B). Frames were extracted and cropped from each examination and an expert gastroenterologist labelled the lesions in each frame. After labelling, the labels were exported as binary images. A deep learning segmentation model (U-Net) was trained with focal Tversky loss as a loss function. Two models were obtained with two different sets of parameters for the loss function. The performance metrics observed were intersection over union and recall and precision.

RESULTS

Analyzing the intersection over union metric, the first segmentation model obtained performed better compared to the second model: 0.8392 (model 1) vs. 0.7990 (model 2). The inference time for both models was between 32.15 milliseconds and 77.59 milliseconds.

CONCLUSIONS

Two segmentation models were obtained in the study. The models performed similarly during training and validation. However, one model was trained to focus on hard-to-predict labels. The proposed segmentation models can represent a first step in automatically extracting time-intensity curves from CEUS examinations.

Learning and visualizing chronic latent representations using electronic health records

Background

Nowadays, patients with chronic diseases such as diabetes and hypertension have reached alarming numbers worldwide. These diseases increase the risk of developing acute complications and involve a substantial economic burden and demand for health resources. The widespread adoption of Electronic Health Records (EHRs) is opening great opportunities for supporting decision-making. Nevertheless, data extracted from EHRs are complex (heterogeneous, high-dimensional and usually noisy), hampering the knowledge extraction with conventional approaches.

Methods

We propose the use of the Denoising Autoencoder (DAE), a Machine Learning (ML) technique allowing to transform high-dimensional data into latent representations (LRs), thus addressing the main challenges with clinical data. We explore in this work how the combination of LRs with a visualization method can be used to map the patient data in a two-dimensional space, gaining knowledge about the distribution of patients with different chronic conditions. Furthermore, this representation can be also used to characterize the patient’s health status evolution, which is of paramount importance in the clinical setting.

Results

To obtain clinical LRs, we considered real-world data extracted from EHRs linked to the University Hospital of Fuenlabrada in Spain. Experimental results showed the great potential of DAEs to identify patients with clinical patterns linked to hypertension, diabetes and multimorbidity. The procedure allowed us to find patients with the same main chronic disease but different clinical characteristics. Thus, we identified two kinds of diabetic patients with differences in their drug therapy (insulin and non-insulin dependant), and also a group of women affected by hypertension and gestational diabetes. We also present a proof of concept for mapping the health status evolution of synthetic patients when considering the most significant diagnoses and drugs associated with chronic patients.

Conclusion

Our results highlighted the value of ML techniques to extract clinical knowledge, supporting the identification of patients with certain chronic conditions. Furthermore, the patient’s health status progression on the two-dimensional space might be used as a tool for clinicians aiming to characterize health conditions and identify their more relevant clinical codes.

Supplementary Information

The online version contains supplementary material available at (10.1186/s13040-022-00303-z).

Deep autoencoder-powered pattern identification of sleep disturbance using multi-site cross-sectional survey data

Pattern identification (PI) is a diagnostic method used in Traditional East Asian medicine (TEAM) to select appropriate and personalized acupuncture points and herbal medicines for individual patients. Developing a reproducible PI model using clinical information is important as it would reflect the actual clinical setting and improve the effectiveness of TEAM treatment. In this paper, we suggest a novel deep learning-based PI model with feature extraction using a deep autoencoder and k-means clustering through a cross-sectional study of sleep disturbance patient data. The data were obtained from an anonymous electronic survey in the Republic of Korea Army (ROKA) members from August 16, 2021, to September 20, 2021. The survey instrument consisted of six sections: demographics, medical history, military duty, sleep-related assessments (Pittsburgh sleep quality index (PSQI), Berlin questionnaire, and sleeping environment), diet/nutrition-related assessments [dietary habit survey questionnaire and nutrition quotient (NQ)], and gastrointestinal-related assessments [gastrointestinal symptom rating scale (GSRS) and Bristol stool scale]. Principal component analysis (PCA) and a deep autoencoder were used to extract features, which were then clustered using the k-means clustering method. The Calinski-Harabasz index, silhouette coefficient, and within-cluster sum of squares were used for internal cluster validation and the final PSQI, Berlin questionnaire, GSRS, and NQ scores were used for external cluster validation. One-way analysis of variance followed by the Tukey test and chi-squared test were used for between-cluster comparisons. Among 4,869 survey responders, 2,579 patients with sleep disturbances were obtained after filtering using a PSQI score of &gt;5. When comparing clustering performance using raw data and extracted features by PCA and the deep autoencoder, the best feature extraction method for clustering was the deep autoencoder (16 nodes for the first and third hidden layers, and two nodes for the second hidden layer). Our model could cluster three different PI types because the optimal number of clusters was determined to be three via the elbow method. After external cluster validation, three PI types were differentiated by changes in sleep quality, dietary habits, and concomitant gastrointestinal symptoms. This model may be applied to the development of artificial intelligence-based clinical decision support systems through electronic medical records and clinical trial protocols for evaluating the effectiveness of TEAM treatment.

Diagnostic Value of Vaginal Ultrasound under Improved Clustering Algorithm Combined with Hysteroscopy in Abnormal Uterine Bleeding

In order to explore the diagnostic value of the improved clustering algorithm of vaginal ultrasound combined with hysteroscopy in abnormal uterine bleeding (AUB), 128 patients diagnosed with AUB in the hospital were selected as the research objects. A K-means improved clustering color image segmentation algorithm was designed and applied to AUB vaginal ultrasound image processing. The running time, mean square error (MSE), and peak to signal noise ratio (PSNR) were calculated to evaluate the algorithm, and the sensitivity, specificity, negative likelihood ratio, and positive likelihood ratio were used to evaluate the diagnostic accuracy of the detection method. In addition, combined with hysteroscopy, a comprehensive evaluation of the diagnostic value of abnormal uterine bleeding diseases was implemented. The results showed that compared with the traditional K-means clustering algorithm, the running time of the improved K-means clustering color image segmentation algorithm in the training set was significantly shortened, the MSE was significantly decreased, and the PSNR was significantly increased ( $P<0.05$ ). The sensitivity, specificity, negative likelihood ratio, and positive likelihood ratio (90.5%, 93.2%, 84.3, and 96.3%) of AUB diagnosis were significantly improved in the algorithm of vaginal ultrasound combined with hysteroscopy ( $P<0.05$ ). In summary, the combination of vaginal ultrasound and hysteroscopy based on K-means improved clustering color image segmentation algorithm can significantly improve the clinical diagnostic accuracy of AUB patients.