Using Innovative Machine Learning Methods to Screen and Identify Predictors of Congenital Heart Diseases

A review of evaluation approaches for explainable AI with applications in cardiology

Explainable artificial intelligence (XAI) elucidates the decision-making process of complex AI models and is important in building trust in model predictions. XAI explanations themselves require evaluation as to accuracy and reasonableness and in the context of use of the underlying AI model. This review details the evaluation of XAI in cardiac AI applications and has found that, of the studies examined, 37% evaluated XAI quality using literature results, 11% used clinicians as domain-experts, 11% used proxies or statistical analysis, with the remaining 43% not assessing the XAI used at all. We aim to inspire additional studies within healthcare, urging researchers not only to apply XAI methods but to systematically assess the resulting explanations, as a step towards developing trustworthy and safe models.

Supplementary Information

The online version contains supplementary material available at 10.1007/s10462-024-10852-w.

Machine Learning Detects Symptomatic Plaques in Patients With Carotid Atherosclerosis on CT Angiography

BACKGROUND

This study aimed to develop and validate a computed tomography angiography based machine learning model that uses plaque composition data and degree of carotid stenosis to detect symptomatic carotid plaques in patients with carotid atherosclerosis.

METHODS

The machine learning based model was trained using degree of stenosis and the volumes of 13 computed tomography angiography derived intracarotid plaque subcomponents (eg, lipid, intraplaque hemorrhage, calcium) to identify plaques associated with cerebrovascular events. The model was internally validated through repeated 10-fold cross-validation and tested on a dedicated testing cohort according to discrimination and calibration.

RESULTS

This retrospective, single-center study evaluated computed tomography angiography scans of 268 patients with both symptomatic and asymptomatic carotid atherosclerosis (163 for the derivation set and 106 for the testing set) performed between March 2013 and October 2019. The area-under-receiver-operating characteristics curve by machine learning on the testing cohort (0.89) was significantly higher than the areas under the curve of traditional logit analysis based on the degree of stenosis (0.51, P<0.001), presence of intraplaque hemorrhage (0.69, P<0.001), and plaque composition (0.78, P<0.001), respectively. Comparable performance was obtained on internal validation. The identified plaque components and associated cutoff values that were significantly associated with a higher likelihood of symptomatic status after adjustment were the ratio of intraplaque hemorrhage to lipid volume (≥50%, 38.5 [10.1-205.1]; odds ratio, 95% CI) and percentage of intraplaque hemorrhage volume (≥10%, 18.5 [5.7-69.4]; odds ratio, 95% CI).

CONCLUSIONS

This study presented an interpretable machine learning model that accurately identifies symptomatic carotid plaques using computed tomography angiography derived plaque composition features, aiding clinical decision-making.

Machine learning detects symptomatic patients with carotid plaques based on 6-type calcium configuration classification on CT angiography

OBJECTIVES

While the link between carotid plaque composition and cerebrovascular vascular (CVE) events is recognized, the role of calcium configuration remains unclear. This study aimed to develop and validate a CT angiography (CTA)-based machine learning (ML) model that uses carotid plaques 6-type calcium grading, and clinical parameters to identify CVE patients with bilateral plaques.

MATERIAL AND METHODS

We conducted a multicenter, retrospective diagnostic study (March 2013-May 2020) approved by the institutional review board. We included adults (18 +) with bilateral carotid artery plaques, symptomatic patients having recently experienced a carotid territory ischemic event, and asymptomatic patients either after 3 months from symptom onset or with no such event. Four ML models (clinical factors, calcium configurations, and both with and without plaque grading [ML-All-G and ML-All-NG]) and logistic regression on all variables identified symptomatic patients. Internal validation assessed discrimination and calibration. External validation was also performed, and identified important variables and causes of misclassifications.

RESULTS

We included 790 patients (median age 72, IQR [61-80], 42% male, 64% symptomatic) for training and internal validation, and 159 patients (age 68 [63-76], 36% male, 39% symptomatic) for external testing. The ML-All-G model achieved an area-under-ROC curve of 0.71 (95% CI 0.58-0.78; p < .001) and sensitivity 80% (79-81). Performance was comparable on external testing. Calcified plaque, especially the positive rim sign on the right artery in older and hyperlipidemic patients, had a major impact on identifying symptomatic patients.

CONCLUSION

The developed model can identify symptomatic patients using plaques calcium configuration data and clinical information with reasonable diagnostic accuracy.

CLINICAL RELEVANCE

The analysis of the type of calcium configuration in carotid plaques into 6 classes, combined with clinical variables, allows for an effective identification of symptomatic patients.

KEY POINTS

• While the association between carotid plaques composition and cerebrovascular events is recognized, the role of calcium configuration remains unclear. • Machine learning of 6-type plaque grading can identify symptomatic patients. Calcified plaques on the right artery, advanced age, and hyperlipidemia were the most important predictors. • Fast acquisition of CTA enables rapid grading of plaques upon the patient's arrival at the hospital, which streamlines the diagnosis of symptoms using ML.

A Cardiac Deep Learning Model (CDLM) to Predict and Identify the Risk Factor of Congenital Heart Disease

Congenital heart disease (CHD) is a critical global public health concern, particularly when it comes to newborn mortality. Low- and middle-income countries face the highest mortality rates due to limited resources and inadequate healthcare access. To address this pressing issue, machine learning presents an opportunity to develop accurate predictive models that can assess the risk of death from CHD. These models can empower healthcare professionals by identifying high-risk infants and enabling appropriate care. Additionally, machine learning can uncover patterns in the risk factors associated with CHD mortality, leading to targeted interventions that prevent or reduce mortality among vulnerable newborns. This paper proposes an innovative machine learning approach to minimize newborn mortality related to CHD. By analyzing data from infants diagnosed with CHD, the model identifies key risk factors contributing to mortality. Armed with this knowledge, healthcare providers can devise customized interventions, including intensified care for high-risk infants and early detection and treatment strategies. The proposed diagnostic model utilizes maternal clinical history and fetal health information to accurately predict the condition of newborns affected by CHD. The results are highly promising, with the proposed Cardiac Deep Learning Model (CDLM) achieving remarkable performance metrics, including a sensitivity of 91.74%, specificity of 92.65%, positive predictive value of 90.85%, negative predictive value of 55.62%, and a miss rate of 91.03%. This research aims to make a significant impact by equipping healthcare professionals with powerful tools to combat CHD-related newborn mortality, ultimately saving lives and improving healthcare outcomes worldwide.

Dietary Inflammatory Index during Pregnancy and Congenital Heart Defects

The relationship between diet-related inflammation during pregnancy and congenital heart defects (CHD) is unclear. This study attempted to investigate the association between the dietary inflammation index (DII) during pregnancy, reflecting the overall inflammatory potential of the maternal diet, and CHD in Northwest China. A case-control study with 474 cases and 948 controls was performed in Xi'an City, China. Eligible women awaiting delivery were recruited, and their dietary and other information during pregnancy was collected. Logistic regression models were applied to estimate the risk of CHD in association with DII. The maternal DII ranged from -1.36 to 5.73 in cases, and 0.43 to 5.63 in controls. Pregnant women with per 1 higher DII score were at 31% higher risk of fetal CHD (OR = 1.31, 95%CI = 1.14-1.51), and the adjusted OR (95%CI) comparing the pro-inflammatory diet group with the anti-inflammatory diet group was 2.04 (1.42-2.92). The inverse association of maternal DII score with CHD risk was consistent across various subgroups of maternal characteristics. Maternal DII in pregnancy had good predictive value for CHD in offspring, with the areas under the receiver operating characteristic curve higher than 0.7. These findings suggested that avoiding a pro-inflammatory diet in pregnancy should be emphasized in the prevention of CHD.

Insights and perspectives into clinical biomarker discovery in pediatric heart failure and congenital heart disease-a narrative review

Background and Objective

Heart failure (HF) in the pediatric population is a multi-factorial process with a wide spectrum of etiologies and clinical manifestations, that are distinct from the adult HF population, with congenital heart disease (CHD) as the most common cause. CHD has high morbidity/mortality with nearly 60% developing HF during the first 12 months of life. Hence, early discovery and diagnosis of CHD in neonates is pivotal. Plasma B-type natriuretic peptide (BNP) is an increasingly popular clinical marker in pediatric HF, however, in contrast to adult HF, it is not yet included in pediatric HF guidelines and there is no standardized reference cut-off value. We explore the current trends and prospects of biomarkers in pediatric HF, including CHD that can aid in diagnosis and management.

Methods

As a narrative review, we will analyze biomarkers with respect to diagnosis and monitoring in specific anatomical types of CHD in the pediatric population considering all English PubMed publications till June 2022.

Key Content and Findings

We present a concise description of our own experience in applying plasma BNP as a clinical biomarker in pediatric HF and CHD (tetralogy of fallot vs. ventricular septal defect) in the context of surgical correction, as well as untargeted metabolomics analyses. In the current age of Information Technology and large data sets we also explored new biomarker discovery using Text Mining of 33M manuscripts currently on PubMed.

Conclusions

(Multi) Omics studies from patient samples as well as Data Mining can be considered for the discovery of potential pediatric HF biomarkers useful in clinical care. Future research should focus on validation and defining evidence-based value limits and reference ranges for specific indications using the most up-to-date assays in parallel to commonly used studies.

Application of explainable artificial intelligence for healthcare: A systematic review of the last decade (2011-2022)

BACKGROUND AND OBJECTIVES

Artificial intelligence (AI) has branched out to various applications in healthcare, such as health services management, predictive medicine, clinical decision-making, and patient data and diagnostics. Although AI models have achieved human-like performance, their use is still limited because they are seen as a black box. This lack of trust remains the main reason for their low use in practice, especially in healthcare. Hence, explainable artificial intelligence (XAI) has been introduced as a technique that can provide confidence in the model's prediction by explaining how the prediction is derived, thereby encouraging the use of AI systems in healthcare. The primary goal of this review is to provide areas of healthcare that require more attention from the XAI research community.

METHODS

Multiple journal databases were thoroughly searched using PRISMA guidelines 2020. Studies that do not appear in Q1 journals, which are highly credible, were excluded.

RESULTS

In this review, we surveyed 99 Q1 articles covering the following XAI techniques: SHAP, LIME, GradCAM, LRP, Fuzzy classifier, EBM, CBR, rule-based systems, and others.

CONCLUSION

We discovered that detecting abnormalities in 1D biosignals and identifying key text in clinical notes are areas that require more attention from the XAI research community. We hope this is review will encourage the development of a holistic cloud system for a smart city.

Clinlabomics: leveraging clinical laboratory data by data mining strategies

The recent global focus on big data in medicine has been associated with the rise of artificial intelligence (AI) in diagnosis and decision-making following recent advances in computer technology. Up to now, AI has been applied to various aspects of medicine, including disease diagnosis, surveillance, treatment, predicting future risk, targeted interventions and understanding of the disease. There have been plenty of successful examples in medicine of using big data, such as radiology and pathology, ophthalmology cardiology and surgery. Combining medicine and AI has become a powerful tool to change health care, and even to change the nature of disease screening in clinical diagnosis. As all we know, clinical laboratories produce large amounts of testing data every day and the clinical laboratory data combined with AI may establish a new diagnosis and treatment has attracted wide attention. At present, a new concept of radiomics has been created for imaging data combined with AI, but a new definition of clinical laboratory data combined with AI has lacked so that many studies in this field cannot be accurately classified. Therefore, we propose a new concept of clinical laboratory omics (Clinlabomics) by combining clinical laboratory medicine and AI. Clinlabomics can use high-throughput methods to extract large amounts of feature data from blood, body fluids, secretions, excreta, and cast clinical laboratory test data. Then using the data statistics, machine learning, and other methods to read more undiscovered information. In this review, we have summarized the application of clinical laboratory data combined with AI in medical fields. Undeniable, the application of Clinlabomics is a method that can assist many fields of medicine but still requires further validation in a multi-center environment and laboratory.

Dietary Quality during Pregnancy and Congenital Heart Defects

Limited studies on maternal dietary quality indices and congenital heart defects (CHD) are available. This study aimed to explore the relationship between dietary quality in pregnancy and CHD among the Chinese population. A case-control study was performed in Northwest China, and 474 cases and 948 controls were included. Eligible women waiting for delivery were interviewed to recall diets and other information during pregnancy. Dietary quality was assessed by the Global Diet Quality Score (GDQS) and Mediterranean Diet Score (MDS). Logistic regression models were adopted to evaluate the associations of dietary quality scores with CHD. Pregnant women with higher scores of GDQS and MDS were at a lower risk of fetal CHD, and the adjusted ORs comparing the extreme quartiles were 0.26 (95%CI: 0.16−0.42; Ptrend < 0.001) and 0.53 (95%CI: 0.34−0.83; Ptrend = 0.007), respectively. The inverse associations of GDQS and MDS with CHD appeared to be stronger among women with lower education levels or in rural areas. Maternal GDQS and MDS had good predictive values for fetal CHD, with the areas under the receiver operating characteristic curves close to 0.8. Efforts to improve maternal dietary quality need to be strengthened to decrease the prevalence of CHD among the Chinese population.

Building a predictive model to identify clinical indicators for COVID-19 using machine learning method

Although some studies tried to identify risk factors for COVID-19, the evidence comparing COVID-19 and community-acquired pneumonia (CAP) is inconclusive, and CAP is the most common pneumonia with similar symptoms as COVID-19. We conducted a case-control study with 35 routine-collected clinical indicators and demographic factors to identify predictors for COVID-19 with CAP as controls. We randomly split the dataset into a training set (70%) and testing set (30%). We built Explainable Boosting Machine to select the important factors and built a decision tree on selected variables to interpret their relationships. The top five individual predictors of COVID-19 are albumin, total bilirubin, monocyte count, alanine aminotransferase, and percentage of monocyte with the importance scores ranging from 0.078 to 0.567. The top systematic predictors for COVID-19 are liver function, monocyte increasing, plasma protein, granulocyte, and renal function (importance scores ranging 0.009-0.096). We identified five combinations of important indicators to screen COVID-19 patients from CAP patients with differentiating abilities ranging 83.3-100%. An online predictive tool for our model was published. Certain clinical indicators collected routinely from most hospitals could help screen and distinguish COVID-19 from CAP. While further verification is needed, our findings and predictive tool could help screen suspected COVID-19 cases.