AI-Driven Biomarker Detection in Medical Diagnostics

Non-invasive monitoring of blood analyte levels using electrocardiograms (ECGs) and machine learning. The technique involves training a machine learning model using ECGs and corresponding analyte levels from multiple individuals. The model can then predict analyte levels from new ECGs without needing blood samples. The training data is processed to filter out potentially inaccurate analyte measurements that could negatively impact model accuracy. This involves analyzing the ECGs to identify features that correlate with analyte levels, as some subtle changes in the ECG may indicate analyte fluctuations that are not obvious to the human eye. The model is trained using these identified ECG features rather than just the analyte measurements. By leveraging the relationship between ECGs and analyte levels, the model can accurately predict analyte concentrations without invasive blood draws.

Source

Health monitoring using multimodal sensors and external data to accurately diagnose diseases and track symptoms over time. The method involves continuously monitoring health using sensors like microphones, peak flow meters, pulse oximeters, etc. It also collects non-sensory data like medication, diet, location, etc. The sensory and non-sensory data is fed into neural networks to detect symptoms and diseases. The networks adapt periodically based on user data to improve accuracy and reduce false alarms. The method also generates personalized audio summaries for doctors to help diagnose.

Source

Machine learning method for detecting periodontal disease using oral microbiome analysis. The method involves training a machine learning model to predict periodontal disease risk based on specific bacteria detected in oral samples. The model is trained on oral microbiome data from patients with and without periodontal disease. The trained model can then be used to evaluate periodontal disease risk in new samples by detecting the selected bacterial species and calculating their proportions. Higher proportions of certain bacteria indicate higher disease risk.

Source

Using graph neural networks to predict disease progression, survival, and therapy response for multi-focal diseases like cancer or COPD based on graph representations of the patient's disease lesions. The method involves applying a trained graph machine learning model to a graph representation of the patient's lesions to generate clinical information for predicting disease outcomes. The graph representation can include both local and global features of the lesions. The trained model considers the full distribution of lesions across organs rather than just isolated lesions. This holistic approach improves prediction accuracy compared to using isolated lesion features.

Source

A system for detecting dielectric changes in matter using microwave scattering, machine learning, and big data to detect diseases like stroke. The system involves collecting microwave scattering data from an array, analyzing it with a learning device, comparing results to imaging, and iteratively improving the learning. The learning uses correlations between microwave and imaging to predict disease states. It leverages big data to learn disease signatures from scattering. This provides a cheap, portable, scalable alternative to traditional imaging for rapid, intelligent diagnostics using microwaves.

Source

A multivariate classification system and method for identifying reliable biomarkers for diseases like hepatocellular carcinoma. The system reduces the dimensionality of biomarker features to improve classification accuracy by calculating effect sizes of differences between groups and using weights and stability metrics to group features. It then extracts weighted features above a threshold as biomarkers for classification. This reduces the number of features and improves robustness compared to direct univariate analysis.

Source

A biomarker identification method using genetic algorithms to find optimal subsets of genes that can predict and diagnose diseases from high-dimensional gene microarray data. The method involves filtering the genes initially using a technique like mRMR, then applying multiple machine learning algorithms to select features. The feature sets from each algorithm are combined and initialized the genetic algorithm population. Improvements like clustering balance global and local search. The genetic algorithm finds the best feature subset for biomarker identification.

Source

Model-based featurization and classifiers for predicting disease states from nucleic acid samples like blood tests. The technique involves using tissue models to determine the tissue origin of sequencing reads from cell-free DNA. The featurization process assigns each read to the tissue model with the highest likelihood of origin. This tissue-based feature vector is then used to train classifiers for detecting diseases. The classifiers generate signal vectors representing disease state detection. By training and testing on labeled samples, cutoff thresholds are determined for each disease state based on the labels. This allows detecting diseases from unknown samples by applying the labeled cutoffs to their signal vectors.

Source

A method for selecting biomarkers for machine learning models to accurately predict treatment response for diseases. The method involves partitioning the set of biomarker genes into subsets, applying multiple machine learning models to each subset to predict treatment response, generating an ensembled feature set of the top performing biomarkers, and then further selecting a final set of biomarkers from the ensembled set based on their impact on the final treatment prediction. This iterative process refines the biomarker selection for more accurate treatment response prediction.

Source

A multi-model feature selection method and system for machine learning applications in medical diagnosis that improves accuracy and reduces computational complexity by selecting relevant features with high contribution. The method involves using multiple machine learning models to analyze feature contribution, assign weights based on model accuracy, calculate feature scores, sort and select the top half as the optimal feature subset. This reduces the number of features while retaining the important ones.

Source

Detecting and monitoring cancer using DNA methylation analysis. The method involves identifying a panel of methylation signatures associated with multiple cancer types. A machine learning model is trained using this panel to distinguish between healthy individuals and those with cancer. By sequencing DNA from a subject and applying the model, cancer detection is possible. The model can also determine the tissue of origin for multi-cancer cases. The method aims to improve early cancer detection with high sensitivity and specificity using a universal panel of cancer-associated methylation signatures.

Source

Automatically detecting and quantifying pathology in medical imaging scans using AI techniques like unsupervised learning to identify and classify irregularities without requiring a human expert. The method involves training a computer system to analyze scanned images and quantify the extent of pathology patterns. It can be used to automatically identify pathologies like usual interstitial pneumonia (UIP) in lung CT scans for diagnosing conditions like idiopathic pulmonary fibrosis. The AI system learns to recognize pathology patterns without human supervision.

Source

A method and system for accurately predicting cervical cancer using deep feature selection to integrate multiple omics data. The method involves using a Lasso (least absolute shrinkage and selection operator) deep feature selection algorithm to extract relevant features from multi-omics data like genomic, epigenomic, transcriptomic, and proteomic data. The Lasso algorithm helps avoid overfitting and improves model generalization by shrinking and selecting important features. The extracted features are then used to train a prediction model for cervical cancer screening. The Lasso deep feature selection enables better accuracy compared to directly using the raw multi-omics data for prediction.

Source

Explainable AI models using CNN and attention for early, non-invasive diagnostics of Alzheimer's Disease by analyzing patient's speech patterns. The models can detect Alzheimer's using two types of features: parts-of-speech (PoS) and language embeddings. Attention layers capture the relative importance of features within each class and between classes. This provides explainability by showing which features are most important for the model's decision-making.

Source

Using machine learning algorithms trained on historical insurance claims data and biomarker levels to identify and monitor diseases like multiple sclerosis before symptoms become apparent. The algorithms can predict disease risk and progression based on patterns in insurance claims data like diagnosis codes, treatments, and demographics. Blood biomarkers like long non-coding RNA levels can also be used. This allows early detection, prognosis, and personalized treatment recommendations for diseases like MS.

Source

Method for diagnosing and staging liver diseases using DNA analysis. The method involves analyzing a DNA sample, like circulating cell-free DNA (cfDNA) or blood cell DNA, to determine methylation patterns at specific CpG sites. Methylation levels are calculated from these patterns and used to classify the liver disease. This allows non-invasive diagnosis and staging of conditions like fatty liver, NASH, cirrhosis, and liver cancer without biopsy. The methylation markers are selected based on tissue specificity and can distinguish between liver and non-liver tissue.

Source

Extracting biomarkers from subject response data like video and audio to determine disease severity. The method involves presenting stimuli to a subject and recording their response. Biomarkers are derived from the recorded data by extracting descriptors like facial expressions, speech patterns, and vocal characteristics. These descriptors are used to calculate biomarkers that indicate disease severity.

Source

Extracting biomarkers for disease diagnosis using a deep neural network from a limited number of samples. The method involves analyzing genomic data from a small set of samples to extract biomarkers for disease diagnosis. The method uses repeated random sampling of the limited data to train and evaluate neural networks. It determines important genomic features that affect network accuracy. By synthesizing features from multiple networks, it ranks and selects biomarkers from the limited data. This allows biomarker extraction using neural networks even with a small number of samples, unlike conventional methods needing large datasets.

Source

A method for identifying colorectal cancer molecular markers using multi-omics data that involves preprocessing the data, training neural network models, and identifying important features. The method aims to improve accuracy of colorectal cancer molecular marker identification compared to single-omics methods. It starts by screening the omics data based on correlation with colorectal cancer. Then, it trains neural network models using the filtered data. Finally, it identifies features from the models that have the greatest impact on classification to find potential colorectal cancer molecular markers.

Source

A blood test-based method using machine learning to diagnose and predict COVID-19 using routine blood test data. The method involves training a machine learning model using blood test data from COVID-19 patients and healthy controls to identify biomarkers that distinguish COVID-19 cases. This model can then be used to predict COVID-19 status from new blood test data. The method aims to provide a cheaper, faster, and more accessible diagnostic alternative to PCR testing, particularly in resource-limited settings where PCR testing may not be feasible.

Source