Search

Abstract Background Cutaneous melanoma is the most aggressive form of skin cancer, responsible for most skin cancer-related deaths. Recent advances in artificial intelligence, jointly with the availability of public dermoscopy image datasets, have allowed to assist dermatologists in melanoma identification. While image feature extraction holds potential for melanoma detection, it often leads to high-dimensional data. Furthermore, most image datasets present the class imbalance problem, where a few classes have numerous samples, whereas others are under-represented. Methods In this paper, we propose to combine ensemble feature selection (FS) methods and data augmentation with the conditional tabular generative adversarial networks (CTGAN) to enhance melanoma identification in imbalanced datasets. We employed dermoscopy images from two public datasets, PH2 and Derm7pt, which contain melanoma and not-melanoma lesions. To capture intrinsic information from skin lesions, we conduct two feature extraction (FE) approaches, including handcrafted and embedding features. For the former, color, geometric and first-, second-, and higher-order texture features were extracted, whereas for the latter, embeddings were obtained using ResNet-based models. To alleviate the high-dimensionality in the FE, ensemble FS with filter methods were used and evaluated. For data augmentation, we conducted a progressive analysis of the imbalance ratio (IR), related to the amount of synthetic samples created, and evaluated the impact on the predictive results. To gain interpretability on predictive models, we used SHAP, bootstrap resampling statistical tests and UMAP visualizations. Results The combination of ensemble FS, CTGAN, and linear models achieved the best predictive results, achieving AUCROC values of 87% (with support vector machine and IR=0.9) and 76% (with LASSO and IR=1.0) for the PH2 and Derm7pt, respectively. We also identified that melanoma lesions were mainly characterized by features related to color, while not-melanoma lesions were characterized by texture features. Conclusions Our results demonstrate the effectiveness of ensemble FS and synthetic data in the development of models that accurately identify melanoma. This research advances skin lesion analysis, contributing to both melanoma detection and the interpretation of main features for its identification.

Noncommunicable diseases (NCDs) are the leading cause of morbidity and mortality worldwide. Cardiovascular diseases (CVDs) and diabetes are the most prevalent NCDs, causing 1.9 and 1.5 million deaths yearly. Individuals diagnosed with type 1 diabetes (T1D) are at high risk of developing CVDs. Machine learning (ML) models have provided outstanding results in different domains, including healthcare, allowing to obtain models with high predictive performance. The aim of this study was to develop an interpretable data-driven approach to predict the 10-year CVD risk for T1D older individuals, aiming to provide both reasonable predictive performance and the identification of risk factors associated with CVDs. Data from T1D individuals at the Steno Diabetes Center Copenhagen were used. Different ML-based models were considered, including KNN, decision tree, random forest, and multilayer perceptron (MLP). To enhance the predictive performance of ML models, the conditional tabular generative adversarial network (CTGAN) was used to create synthetic data and increase the size of the training data. Several filter and wrapper feature selection (FS) techniques were considered for identifying the most relevant features involved in CVD risk and enhancing the performance of the ML-based models used. To gain interpretability on predictive models, we used the post-hoc methods: SHAP and accumulated local effects. The experimental results showed a great performance of FS and ML-based models for predicting CVD risk. In particular, the MLP obtained the best results, with a mean absolute error of 0.0088 and mean relative absolute error of 0.0817. Regarding risk factors, age, Hba1c, and albuminuria were identified as crucial in CVD risk prediction, which is in line with recent clinical evidence. Our study contributes to identifying CVD risk and associated risk factors in a data-driven manner, helping to make early interventions and adequate treatments to prevent CVDs.

Continuous glucose monitoring (CGM) represents a significant advancement in diabetes management, playing an important role in glycemic control for patients with type 1 diabetes (T1D). Despite their benefits, their performance is affected by numerous factors such as the carbohydrate intake, alcohol consumption, and physical activity (PA). Among these, PA could cause hypoglycemic episodes, which might happen after exercising. In this work, two main contributions are presented. First, we extend the performance evaluation of two glucose monitoring devices, Eversense and Free Style Libre (FSL), for measuring glucose concentrations during high-intensity PA and normal daily activity (NDA). The impact of PA is investigated considering (1) different glucose ranges (hypoglycemia, euglycemia, and hyperglycemia); and (2) four time periods throughout the day (morning, afternoon, evening, and night). Second, we evaluate the effectiveness of machine learning (ML) models, including logistic regression, K-nearest neighbors, and support vector machine, to automatically detect PA in T1D individuals using glucose measurements. The performance analysis showed significant differences between glucose levels obtained in the PA and NDA period for Eversense and FSL devices, specially in the hyperglycemic range and two time intervals (morning and afternoon). Both Eversense and FSL devices present measurements with large variability during strenuous PA, indicating that their users should be cautious. However, glucose recordings provided by monitoring devices are accurate for NDA, reaching similar values to capillary glucose device. Lastly, ML-based models yielded promising results to determine when an individual has performed PA, reaching an accuracy value of 0.93. The results can be used to develop an individualized data-driven classifier for each patient that categorizes glucose profiles based on the time interval during the day and according to if a patient performs PA. Our work contributes to the analysis of PA on the performance of CGM devices.

Abstract 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.

Machine Learning (ML) methods have become important for enhancing the performance of decision-support predictive models. However, class imbalance is one of the main challenges for developing ML models, because it may bias the learning process and the model generalization ability. In this paper, we consider oversampling methods for generating synthetic categorical clinical data aiming to improve the predictive performance in ML models, and the identification of risk factors for cardiovascular diseases (CVDs). We performed a comparative study of several categorical synthetic data generation methods, including Synthetic Minority Oversampling Technique Nominal (SMOTEN), Tabular Variational Autoencoder (TVAE) and Conditional Tabular Generative Adversarial Networks (CTGANs). Then, we assessed the impact of combining oversampling strategies and linear and nonlinear supervised ML methods. Lastly, we conducted a post-hoc model interpretability based on the importance of the risk factors. Experimental results show the potential of GAN-based models for generating high-quality categorical synthetic data, yielding probability mass functions that are very close to those provided by real data, maintaining relevant insights, and contributing to increasing the predictive performance. The GAN-based model and a linear classifier outperform other oversampling techniques, improving the area under the curve by 2%. These results demonstrate the capability of synthetic data to help with both determining risk factors and building models for CVD prediction.

Hypertension and diabetes have become a global health and economic issue, being among the major chronic conditions worldwide, particularly in developed countries. To face this global problem, a better knowledge about these diseases becomes crucial to characterize chronic patients. Our aim is two-fold: (1) to provide an efficient visual tool for identifying clinical patterns in high-dimensional data; and (2) to characterize the patient health-status through a data-driven approach using electronic health records of healthy, hypertensive and diabetic populations. We propose a two-stage methodology that uses diagnosis and drug codes of healthy and chronic patients associated to the University Hospital of Fuenlabrada in Spain. The first stage applies the Self-Organizing Map on the aforementioned data to get a set of prototype patients which are projected onto a grid of nodes. Each node has associated a prototype patient that captures relationships among clinical characteristics. In the second stage, clustering methods are applied on the prototype patients to find groups of patients with a similar health-status. Clusters with distinctive patterns linked to chronic conditions were found, being the most remarkable highlights: a cluster of pregnant women emerged among the hypertensive population, and two clusters of diabetic individuals with significant differences in drug-therapy (insulin and non-insulin dependant). The proposed methodology showed to be effective to explore relationships within clinical data and to find patterns related to diabetes and hypertension in a visual way. Our methodology raises as a suitable alternative for building appropriate clinical groups, becoming a promising approach to be applied to any population due to its data-driven philosophy. A thorough analysis of these groups could spawn new and fruitful findings.