Unsupervised machine learning and prognostic factors of survival in chronic lymphocytic leukemia

SillyPutty: Improved clustering by optimizing the silhouette width

Clustering is an important task in biomedical science, and it is widely believed that different data sets are best clustered using different algorithms. When choosing between clustering algorithms on the same data set, reseachers typically rely on global measures of quality, such as the mean silhouette width, and overlook the fine details of clustering. However, the silhouette width actually computes scores that describe how well each individual element is clustered. Inspired by this observation, we developed a novel clustering method, called SillyPutty. Unlike existing methods, SillyPutty uses the silhouette width for individual elements as a tool to optimize the mean silhouette width. This shift in perspective allows for a more granular evaluation of clustering quality, potentially addressing limitations in current methodologies. To test the SillyPutty algorithm, we first simulated a series of data sets using the Umpire R package and then used real-workd data from The Cancer Genome Atlas. Using these data sets, we compared SillyPutty to several existing algorithms using multiple metrics (Silhouette Width, Adjusted Rand Index, Entropy, Normalized Within-group Sum of Square errors, and Perfect Classification Count). Our findings revealed that SillyPutty is a valid standalone clustering method, comparable in accuracy to the best existing methods. We also found that the combination of hierarchical clustering followed by SillyPutty has the best overall performance in terms of both accuracy and speed. Availability: The SillyPutty R package can be downloaded from the Comprehensive R Archive Network (CRAN).

SillyPutty: Improved clustering by optimizing the silhouette width

Unsupervised clustering is an important task in biomedical science. We developed a new clustering method, called SillyPutty, for unsupervised clustering. As test data, we generated a series of datasets using the Umpire R package. Using these datasets, we compared SillyPutty to several existing algorithms using multiple metrics (Silhouette Width, Adjusted Rand Index, Entropy, Normalized Within-group Sum of Square errors, and Perfect Classification Count). Our findings revealed that SillyPutty is a valid standalone clustering method, comparable in accuracy to the best existing methods. We also found that the combination of hierarchical clustering followed by SillyPutty has the best overall performance in terms of both accuracy and speed.

Clinical Informatics Approaches to Understand and Address Cancer Disparities

OBJECTIVES

Disparities in cancer incidence and outcomes across race, ethnicity, gender, socioeconomic status, and geography are well-documented, but their etiologies are often poorly understood and multifactorial. Clinical informatics can provide tools to better understand and address these disparities by enabling high-throughput analysis of multiple types of data. Here, we review recent efforts in clinical informatics to study and measure disparities in cancer.

METHODS

We carried out a narrative review of clinical informatics studies related to cancer disparities and bias published from 2018-2021, with a focus on domains such as real-world data (RWD) analysis, natural language processing (NLP), radiomics, genomics, proteomics, metabolomics, and metagenomics.

RESULTS

Clinical informatics studies that investigated cancer disparities across race, ethnicity, gender, and age were identified. Most cancer disparities work within clinical informatics used RWD analysis, NLP, radiomics, and genomics. Emerging applications of clinical informatics to understand cancer disparities, including proteomics, metabolomics, and metagenomics, were less well represented in the literature but are promising future research avenues. Algorithmic bias was identified as an important consideration when developing and implementing cancer clinical informatics techniques, and efforts to address this bias were reviewed.

CONCLUSIONS

In recent years, clinical informatics has been used to probe a range of data sources to understand cancer disparities across different populations. As informatics tools become integrated into clinical decision-making, attention will need to be paid to ensure that algorithmic bias does not amplify existing disparities. In our increasingly interconnected medical systems, clinical informatics is poised to untap the full potential of multi-platform health data to address cancer disparities.

Intersubject Variability in Cerebrovascular Hemodynamics and Systemic Physiology during a Verbal Fluency Task under Colored Light Exposure: Clustering of Subjects by Unsupervised Machine Learning

There is large intersubject variability in cerebrovascular hemodynamic and systemic physiological responses induced by a verbal fluency task (VFT) under colored light exposure (CLE). We hypothesized that machine learning would enable us to classify the response patterns and provide new insights into the common response patterns between subjects. In total, 32 healthy subjects (15 men and 17 women, age: 25.5 ± 4.3 years) were exposed to two different light colors (red vs. blue) in a randomized cross-over study design for 9 min while performing a VFT. We used the systemic physiology augmented functional near-infrared spectroscopy (SPA-fNIRS) approach to measure cerebrovascular hemodynamics and oxygenation at the prefrontal cortex (PFC) and visual cortex (VC) concurrently with systemic physiological parameters. We found that subjects were suitably classified by unsupervised machine learning into different groups according to the changes in the following parameters: end-tidal carbon dioxide, arterial oxygen saturation, skin conductance, oxygenated hemoglobin in the VC, and deoxygenated hemoglobin in the PFC. With hard clustering methods, three and five different groups of subjects were found for the blue and red light exposure, respectively. Our results highlight the fact that humans show specific reactivity types to the CLE-VFT experimental paradigm.

Comparing machine learning algorithms to predict 5-year survival in patients with chronic myeloid leukemia

Introduction

Chronic myeloid leukemia (CML) is a myeloproliferative disorder resulting from the translocation of chromosomes 19 and 22. CML includes 15–20% of all cases of leukemia. Although bone marrow transplant and, more recently, tyrosine kinase inhibitors (TKIs) as a first-line treatment have significantly prolonged survival in CML patients, accurate prediction using available patient-level factors can be challenging. We intended to predict 5-year survival among CML patients via eight machine learning (ML) algorithms and compare their performance.

Methods

The data of 837 CML patients were retrospectively extracted and randomly split into training and test segments (70:30 ratio). The outcome variable was 5-year survival with potential values of alive or deceased. The dataset for the full features and important features selected by minimal redundancy maximal relevance (mRMR) feature selection were fed into eight ML techniques, including eXtreme gradient boosting (XGBoost), multilayer perceptron (MLP), pattern recognition network, k-nearest neighborhood (KNN), probabilistic neural network, support vector machine (SVM) (kernel = linear), SVM (kernel = RBF), and J-48. The scikit-learn library in Python was used to implement the models. Finally, the performance of the developed models was measured using some evaluation criteria with 95% confidence intervals (CI).

Results

Spleen palpable, age, and unexplained hemorrhage were identified as the top three effective features affecting CML 5-year survival. The performance of ML models using the selected-features was superior to that of the full-features dataset. Among the eight ML algorithms, SVM (kernel = RBF) had the best performance in tenfold cross-validation with an accuracy of 85.7%, specificity of 85%, sensitivity of 86%, F-measure of 87%, kappa statistic of 86.1%, and area under the curve (AUC) of 85% for the selected-features. Using the full-features dataset yielded an accuracy of 69.7%, specificity of 69.1%, sensitivity of 71.3%, F-measure of 72%, kappa statistic of 75.2%, and AUC of 70.1%.

Conclusions

Accurate prediction of the survival likelihood of CML patients can inform caregivers to promote patient prognostication and choose the best possible treatment path. While external validation is required, our developed models will offer customized treatment and may guide the prescription of personalized medicine for CML patients.

A Review of Artificial Intelligence Applications in Hematology Management: Current Practices and Future Prospects

Background

Machine learning (ML) and deep learning (DL) methods have recently garnered a great deal of attention in the field of cancer research by making a noticeable contribution to the growth of predictive medicine and modern oncological practices. Considerable focus has been particularly directed toward hematologic malignancies because of the complexity in detecting early symptoms. Many patients with blood cancer do not get properly diagnosed until their cancer has reached an advanced stage with limited treatment prospects. Hence, the state-of-the-art revolves around the latest artificial intelligence (AI) applications in hematology management.

Objective

This comprehensive review provides an in-depth analysis of the current AI practices in the field of hematology. Our objective is to explore the ML and DL applications in blood cancer research, with a special focus on the type of hematologic malignancies and the patient’s cancer stage to determine future research directions in blood cancer.

Methods

We searched a set of recognized databases (Scopus, Springer, and Web of Science) using a selected number of keywords. We included studies written in English and published between 2015 and 2021. For each study, we identified the ML and DL techniques used and highlighted the performance of each model.

Results

Using the aforementioned inclusion criteria, the search resulted in 567 papers, of which 144 were selected for review.

Conclusions

The current literature suggests that the application of AI in the field of hematology has generated impressive results in the screening, diagnosis, and treatment stages. Nevertheless, optimizing the patient’s pathway to treatment requires a prior prediction of the malignancy based on the patient’s symptoms or blood records, which is an area that has still not been properly investigated.

A deep learning method and device for bone marrow imaging cell detection

BACKGROUND

Morphological analysis of bone marrow cells is considered as the gold standard for the diagnosis of leukemia. However, due to the diverse morphology of bone marrow cells, extensive experience and patience are needed for morphological examination. automatic diagnosis system through the comprehensive application of image analysis and pattern recognition technology is urgently needed to reduce work intensity, error probability and improves work efficiency.

METHODS

In this article, we establish a new morphological diagnosis system for bone marrow cell detection based on the deep learning object detection framework. The model is based on the Faster Region-Convolutional Neural Network (R-CNN), a classical object detection model. The system automatically detects bone marrow cells and determines their types. As specimens have severe long-tail distribution, i.e., the frequency of different types of cells varies dramatically, we proposed a general score ranking loss to solve such a problem. The general score ranking loss considers the ranking relationship between positive and negative samples and optimizes the positive sample with a higher classification probability value.

RESULTS

We verified this system with 70 bone marrow specimens of leukemia patients, which proved that it can realize intelligent recognition with high efficiency. The software is finally integrated into the microscope system to build an augmented reality system.

CONCLUSIONS

Clinical tests show that the response speed of the newly developed diagnostic system is faster than that of trained diagnostic experts.

Simulation-derived best practices for clustering clinical data

INTRODUCTION

Clustering analyses in clinical contexts hold promise to improve the understanding of patient phenotype and disease course in chronic and acute clinical medicine. However, work remains to ensure that solutions are rigorous, valid, and reproducible. In this paper, we evaluate best practices for dissimilarity matrix calculation and clustering on mixed-type, clinical data.

METHODS

We simulate clinical data to represent problems in clinical trials, cohort studies, and EHR data, including single-type datasets (binary, continuous, categorical) and 4 data mixtures. We test 5 single distance metrics (Jaccard, Hamming, Gower, Manhattan, Euclidean) and 3 mixed distance metrics (DAISY, Supersom, and Mercator) with 3 clustering algorithms (hierarchical (HC), k-medoids, self-organizing maps (SOM)). We quantitatively and visually validate by Adjusted Rand Index (ARI) and silhouette width (SW). We applied our best methods to two real-world data sets: (1) 21 features collected on 247 patients with chronic lymphocytic leukemia, and (2) 40 features collected on 6000 patients admitted to an intensive care unit.

RESULTS

HC outperformed k-medoids and SOM by ARI across data types. DAISY produced the highest mean ARI for mixed data types for all mixtures except unbalanced mixtures dominated by continuous data. Compared to other methods, DAISY with HC uncovered superior, separable clusters in both real-world data sets.

DISCUSSION

Selecting an appropriate mixed-type metric allows the investigator to obtain optimal separation of patient clusters and get maximum use of their data. Superior metrics for mixed-type data handle multiple data types using multiple, type-focused distances. Better subclassification of disease opens avenues for targeted treatments, precision medicine, clinical decision support, and improved patient outcomes.

A novel model to label delirium in an intensive care unit from clinician actions

BACKGROUND

In the intensive care unit (ICU), delirium is a common, acute, confusional state associated with high risk for short- and long-term morbidity and mortality. Machine learning (ML) has promise to address research priorities and improve delirium outcomes. However, due to clinical and billing conventions, delirium is often inconsistently or incompletely labeled in electronic health record (EHR) datasets. Here, we identify clinical actions abstracted from clinical guidelines in electronic health records (EHR) data that indicate risk of delirium among intensive care unit (ICU) patients. We develop a novel prediction model to label patients with delirium based on a large data set and assess model performance.

METHODS

EHR data on 48,451 admissions from 2001 to 2012, available through Medical Information Mart for Intensive Care-III database (MIMIC-III), was used to identify features to develop our prediction models. Five binary ML classification models (Logistic Regression; Classification and Regression Trees; Random Forests; Naïve Bayes; and Support Vector Machines) were fit and ranked by Area Under the Curve (AUC) scores. We compared our best model with two models previously proposed in the literature for goodness of fit, precision, and through biological validation.

RESULTS

Our best performing model with threshold reclassification for predicting delirium was based on a multiple logistic regression using the 31 clinical actions (AUC 0.83). Our model out performed other proposed models by biological validation on clinically meaningful, delirium-associated outcomes.

CONCLUSIONS

Hurdles in identifying accurate labels in large-scale datasets limit clinical applications of ML in delirium. We developed a novel labeling model for delirium in the ICU using a large, public data set. By using guideline-directed clinical actions independent from risk factors, treatments, and outcomes as model predictors, our classifier could be used as a delirium label for future clinically targeted models.

Umpire 2.0: Simulating realistic, mixed-type, clinical data for machine learning

The Umpire 2.0 R-package offers a streamlined, user-friendly workflow to simulate complex, heterogeneous, mixed-type data with known subgroup identities, dichotomous outcomes, and time-to-event data, while providing ample opportunities for fine-tuning and flexibility. Here, we describe how we have expanded the core Umpire 1.0 R-package, developed to simulate gene expression data, to generate clinically realistic, mixed-type data for use in evaluating unsupervised and supervised machine learning (ML) methods. As the availability of large-scale clinical data for ML has increased, clinical data has posed unique challenges, including widely variable size, individual biological heterogeneity, data collection and measurement noise, and mixed data types. Developing and validating ML methods for clinical data requires data sets with known ground truth, generated from simulation. Umpire 2.0 addresses challenges to simulating realistic clinical data by providing the user a series of modules to generate survival parameters and subgroups, apply meaningful additive noise, and discretize to single or mixed data types. Umpire 2.0 provides broad functionality across sample sizes, feature spaces, and data types, allowing the user to simulate correlated, heterogeneous, binary, continuous, categorical, or mixed type data from the scale of a small clinical trial to data on thousands of patients drawn from electronic health records. The user may generate elaborate simulations by varying parameters in order to compare algorithms or interrogate operating characteristics of an algorithm in both supervised and unsupervised ML.