Applications of Neural Networks in Biomedical Data Analysis

Clinical and Biochemical Data for the Diagnosis of Endogenous Hypercortisolism: The "Cushingomic" Approach

CONTEXT

The clinical presentation of Cushing syndrome (CS) overlaps with common conditions. Recommended screening tests are serum cortisol after 1-mg overnight dexamethasone suppression test (DST), urinary free cortisol (UFC), and late-night salivary cortisol (LNSC).

OBJECTIVE

We analyzed the diagnostic accuracy of screening tests in 615 patients without CS (263 suspected CS, 319 adrenal and 33 pituitary incidentaloma) and 40 with CS.

METHODS

Principal component analysis, K-means clustering, and neural network were used to compute an integrated analysis among tests, comorbidities, and signs/symptoms of hypercortisolism.

RESULTS

The diagnostic accuracy of screening tests for CS was high; DST and UFC were slightly superior to LNSC. The threshold of DST should be adapted to the population considered, especially in adrenal incidentaloma (AI) with mild autonomous cortisol secretion: The cutoff to differentiate CS should be increased to 196 nmol/L. Diabetes, hypertension, and obesity were more common in patients without CS: The direction of their vectors was not aligned and their correlation with screening tests was poor. Clustering allowed us to differentiate those patients without CS into cluster 1 (aged osteoporotic patients with impaired screening tests), cluster 2 (hypertensive and metabolic phenotype), and cluster 3 (young individuals with a low likelihood of overt CS). A neural network model that combined screening tests and clinical presentation was able to predict the CS diagnosis in the validation cohort with 99% precision and 86% accuracy.

CONCLUSION

Despite the high diagnostic accuracy of screening tests to detect CS, cortisol-related comorbidities or AI should be considered when interpreting a positive test.

Techniques for learning and transferring knowledge for microbiome-based classification and prediction: review and assessment

The volume of microbiome data is growing at an exponential rate, and the current methodologies for big data mining are encountering substantial obstacles. Effectively managing and extracting valuable insights from these vast microbiome datasets has emerged as a significant challenge in the field of contemporary microbiome research. This comprehensive review delves into the utilization of foundation models and transfer learning techniques within the context of microbiome-based classification and prediction tasks, advocating for a transition away from traditional task-specific or scenario-specific models towards more adaptable, continuous learning models. The article underscores the practicality and benefits of initially constructing a robust foundation model, which can then be fine-tuned using transfer learning to tackle specific context tasks. In real-world scenarios, the application of transfer learning empowers models to leverage disease-related data from one geographical area and enhance diagnostic precision in different regions. This transition from relying on "good models" to embracing "adaptive models" resonates with the philosophy of "teaching a man to fish" thereby paving the way for advancements in personalized medicine and accurate diagnosis. Empirical research suggests that the integration of foundation models with transfer learning methodologies substantially boosts the performance of models when dealing with large-scale and diverse microbiome datasets, effectively mitigating the challenges posed by data heterogeneity.

Exploring the association between early childhood caries, malnutrition, and anemia by machine learning algorithm

OBJECTIVE

The objective of this study was to determine the prevalence of early childhood caries in children with severe acute malnutrition (SAM) and also the hierarchy of association if any with malnutrition, anemia, and other risk factors with ECC using machine learning algorithms.

METHODS

A hospital-based preventive and interventional study was conducted on SAM children (age = 2 to <6 years) who were admitted to the malnutrition treatment unit (MTU). An oral examination for early childhood caries status was done using the deft index. The anthropometric measurements and blood examination reports were recorded. Oral health education and preventive dental treatments were given to the admitted children. Three machine learning algorithms (Random Tree, CART, and Neural Network) were applied to assess the relationship between early childhood caries, malnutrition, anemia, and the risk factors.

RESULTS

The Random Tree model showed that age was the most significant factor in predicting ECC with predictor importance of 98.75%, followed by maternal education (29.20%), hemoglobin level (16.67%), frequency of snack intake (9.17%), deft score (8.75%), consumption of snacks (7.1%), breastfeeding (6.25%), severe acute malnutrition (5.42%), frequency of sugar intake (3.75%), and religion at the minimum predictor importance of 2.08%.

CONCLUSION

Anemia and malnutrition play a significant role in the prediction, hence in the causation of ECC. Pediatricians should also keep in mind that anemia and malnutrition have a negative impact on children's dental health. Hence, Pediatricians and Pediatric dentist should work together in treating this health problem.

The Role of Neural Network Analysis in Identifying Predictors of Gastric Cancer

Background

Gastric cancer is one of the most common cancers. We can use AI for predictive models and help us in early detection and diagnosis.

Objective

This study examines the use of a neural network model to classify gastric cancer based on clinical, demographic and genetic data.

Methods

The data from the participants were divided into two subsets. 70% training data and 30% testing data. The neural network model has 12 input variables. Factors influencing a disease can be age, sex, family history, smoking, alcohol, Helicobacter pylori infection, food habits, diseases, endoscopic images, biopsy, CT scan, gene variants (TP53, KRAS, CDH1). The hyperbolic tangent activation function has four units in the hidden layer of a model. The output layer used a Softmax activation function and cross-entropy error function which predicted the presence of gastric cancer. The assessment was done on the predictors.

Results

The training and testing datasets showed 100% accuracy predicting gastric cancer in the model outputs. Age, gender, family history, infection with Helicobacter pylori, smoking, and drinking alcohol are the biggest predictors. Information from clinical diagnosis like endoscopic images, biopsy and CT scans helped the predictive model.

Conclusion

The neural network was able to perform well for gastric cancer predictions using multiple clinical and demographic factors, showing great utility. The outcomes for AI-based diagnostic tools look promising in cancer, however generalization needs to be confirmed using external datasets. The study shows how artificial intelligence can better precision medicine and cancer diagnosis.

Why Does Rehabilitation Not (Always) Work in Osteoarthritis? Does Rehabilitation Need Molecular Biology?

Osteoarthritis (OA) is a common disease among the human population worldwide. OA causes functional impairment, leads to disability and poses serious socioeconomic burden. The rehabilitation offers a function-oriented method to reduce the disability using diverse interventions (kinesiotherapy, physical therapy, occupational therapy, education, and pharmacotherapy). OA as a widespread disease among elderly patients is often treated by rehabilitation specialists and physiotherapists, however the results of rehabilitation are sometimes unsatisfactory. The understanding of molecular mechanisms activated by rehabilitation may enable the development of more effective rehabilitation procedures. Molecular biology methods may prove crucial in rehabilitation as the majority of rehabilitation procedures cannot be estimated in double-blinded placebo-controlled trials commonly used in pharmacotherapy. This article attempts to present and estimate the role of molecular biology in the development of modern rehabilitation. The role of clinicians in adequate molecular biology experimental design is also described.

Artificial neural networks in contemporary toxicology research

Artificial neural networks (ANNs) have a huge potential in toxicology research. They may be used to predict toxicity of various chemical compounds or classify the compounds based on their toxic effects. Today, numerous ANN models have been developed, some of which may be used to detect and possibly explain complex chemico-biological interactions. Fully connected multilayer perceptrons may in some circumstances have high classification accuracy and discriminatory power in separating damaged from intact cells after exposure to a toxic substance. Regularized and not fully connected convolutional neural networks can detect and identify discrete changes in patterns of two-dimensional data associated with toxicity. Bayesian neural networks with weight marginalization sometimes may have better prediction performance when compared to traditional approaches. With the further development of artificial intelligence, it is expected that ANNs will in the future become important parts of various accurate and affordable biosensors for detection of various toxic substances and evaluation of their biochemical properties. In this concise review article, we discuss the recent research focused on the scientific value of ANNs in evaluation and prediction of toxicity of chemical compounds.