Objective: To classify lung (LS) and heart (HS) sounds between healthy and diseased subjects using a GMM-HMM model. Theoretical Framework: Feature extractors highlight class differences, in this case two types of acoustic vectors were used: MFCC and quantiles. In addition, it is important to determine the number of clusters in the data, in order to use that information when applying the GMM-HMM model. For this, dendrograms, silhouettes and the Bayesian information criterion graph were used. Finally, the model was generated with the combination of a Gaussian mixture model (GMM) and a hidden Markov model (HMM). A GMM model is a triplet composed of the means, covariances and weights. Generally, each HMM state is represented by a GMM model. An HMM can be seen as a model based on states that are not directly observed, in fact, each state is characterized by a GMM. Methods: This paper presents a methodology to optimize the use of Hidden Markov Models and Gaussian Mixture Models to classify lung sounds (LS) and heart sounds (HS) as part of a cardiopulmonary diagnostic evaluation. Specifically, this method was applied as a criterion to choose the best number of clusters in the calculation of a GMM-HMM model. To achieve this, a representation of LS and HS signals in terms of mel frequency cepstral coefficient (MFCC) acoustic vectors and quantile vectors was applied. The analysis of the number of clusters is obtained from dendrograms, silhouettes and the Bayesian information criterion (BIC). Results and Discussion: The GMM-HMM models for LS signals with quartiles yielded excellent results, while for HS signals, the best results were obtained with MFCC vectors. In both cases, for LS and HS signals, a high degree of efficiency was achieved, reaching 100% classification for the set of signals used. The results obtained show the potential of applying techniques such as BIC or dendrograms as a criterion prior to the design of a GMM-HMM architecture. Research implications: This study underlines the importance of a method that helps physicians diagnose/classify LS and HS sounds from healthy and sick subjects. Originality/Value: This study provides a model based on two techniques that allows the classification of LS and HS signals, emphasizing the analysis of clusters.

Objective: To classify healthy subjects from subjects with osteoporosis through X-ray images using Gaussian mixture models (GMM). Theoretical Framework: Osteoporosis is a disease that affects a large number of people, causing fractures of the hip and the thoracic spine. Strong bones primarily reflect the integration of bone mineral density (BMD) and bone quality. There are several techniques available for noninvasive measurements of bone mineral density (BMD) or bone mass. Methods: In this work, a method based on statistical methods for the treatment of bone images is proposed. Particularly, quartile vectors, principal component analysis (PCA) and Gaussian mixture models (GMM) prove to be effective for making statistical computerized diagnoses, determining the presence of osteoporosis. Preprocessing is applied to the X-ray images to standardize all the images, then the images were segmented and their histogram and subsequently their quartile vectors are calculated. PCA is applied to these vectors and a 4-Gaussian GMM model is generated with the resulting vectors. Results and Discussion: The use of PCA on the quartile vectors reduced their dimension and generated better feature vectors, managing to calculate the models and classify the images efficiently. The classification result of the 4-Gaussian GMM model was 100% efficient. Research implications: This study highlights the importance of classifying (diagnosing) osteoporosis by other methods. Originality/Value: This study contributes to the detection of osteoporosis by a novel method that is computationally lower than some other machine learning tools.