Develop Machine Learning Algorithms Research Articles

Machine learning in the prediction of immunotherapy response and prognosis of melanoma: a systematic review and meta-analysis

BackgroundThe emergence of immunotherapy has changed the treatment modality for melanoma and prolonged the survival of many patients. However, a handful of patients remain unresponsive to immunotherapy and effective tools for early identification of this patient population are still lacking. Researchers have developed machine learning algorithms for predicting immunotherapy response in melanoma, but their predictive accuracy has been inconsistent. Therefore, the present systematic review and meta-analysis was performed to comprehensively evaluate the predictive accuracy of machine learning in melanoma response to immunotherapy.MethodsRelevant studies were searched in PubMed, Web of Sciences, Cochrane Library, and Embase from their inception to July 30, 2022. The risk of bias and applicability of the included studies were assessed using the Prediction Model Risk of Bias Assessment Tool (PROBAST). Meta-analysis was performed on R4.2.0.ResultsA total of 36 studies consisting of 30 cohort studies and 6 case-control studies were included. These studies were mainly published between 2019 and 2022 and encompassed 75 models. The outcome measures of this study were progression-free survival (PFS), overall survival (OS), and treatment response. The pooled c-index was 0.728 (95%CI: 0.629–0.828) for PFS in the training set, 0.760 (95%CI: 0.728–0.792) and 0.819 (95%CI: 0.757–0.880) for treatment response in the training and validation sets, respectively, and 0.746 (95%CI: 0.721–0.771) and 0.700 (95%CI: 0.677–0.724) for OS in the training and validation sets, respectively.ConclusionMachine learning has considerable predictive accuracy in melanoma immunotherapy response and prognosis, especially in the former. However, due to the lack of external validation and the scarcity of certain types of models, further studies are warranted.

-A machine learning model to predict surgical site infection after surgery of lower extremity fractures.

This study aimed to develop machine learning algorithms for identifying predictive factors associated with the risk of postoperative surgical site infection in patients with lower extremity fractures. A machine learning analysis was conducted on a dataset comprising 1,579 patients who underwent surgical fixation for lower extremity fractures to create a predictive model for risk stratification of postoperative surgical site infection. We evaluated different clinical and demographic variables to train four machine learning models (neural networks, boosted generalised linear model, naïve bayes, and penalised discriminant analysis). Performance was measured by the area under the curve score, Youdon's index and Brier score. A multivariate adaptive regression splines (MARS) was used to optimise predictor selection. The final model consisted of five predictors. (1) Operating room time, (2) ankle region, (3) open injury, (4) body mass index, and (5) age. The best-performing machine learning algorithm demonstrated a promising predictive performance, with an area under the ROC curve, Youdon's index, and Brier score of 77.8%, 62.5%, and 5.1%-5.6%, respectively. The proposed predictive model not only assists surgeons in determining high-risk factors for surgical site infections but also empowers patients to closely monitor these factors and take proactive measures to prevent complications. Furthermore, by considering the identified predictors, this model can serve as a reference for implementing preventive measures and reducing postoperative complications, ultimately enhancing patient outcomes. However, further investigations involving larger datasets and external validations are required to confirm the reliability and applicability of our model.

Machine learning for predicting Chagas disease infection in rural areas of Brazil.

Chagas disease is a severe parasitic illness that is prevalent in Latin America and often goes unaddressed. Early detection and treatment are critical in preventing the progression of the illness and its associated life-threatening complications. In recent years, machine learning algorithms have emerged as powerful tools for disease prediction and diagnosis. In this study, we developed machine learning algorithms to predict the risk of Chagas disease based on five general factors: age, gender, history of living in a mud or wooden house, history of being bitten by a triatomine bug, and family history of Chagas disease. We analyzed data from the Retrovirus Epidemiology Donor Study (REDS) to train five popular machine learning algorithms. The sample comprised 2,006 patients, divided into 75% for training and 25% for testing algorithm performance. We evaluated the model performance using precision, recall, and AUC-ROC metrics. The Adaboost algorithm yielded an AUC-ROC of 0.772, a precision of 0.199, and a recall of 0.612. We simulated the decision boundary using various thresholds and observed that in this dataset a threshold of 0.45 resulted in a 100% recall. This finding suggests that employing such a threshold could potentially save 22.5% of the cost associated with mass testing of Chagas disease. Our findings highlight the potential of applying machine learning to improve the sensitivity and effectiveness of Chagas disease diagnosis and prevention. Furthermore, we emphasize the importance of integrating socio-demographic and environmental factors into neglected disease prediction models to enhance their performance.

Generalizability of machine learning models predicting 30-day unplanned readmission after primary total knee arthroplasty using a nationally representative database.

Unplanned readmission after primary total knee arthroplasty (TKA) costs an average of US $39,000 per episode and negatively impacts patient outcomes. Although predictive machine learning (ML) models show promise for risk stratification in specific populations, existing studies do not address model generalizability. This study aimed to establish the generalizability of previous institutionally developed ML models to predict 30-day readmission following primary TKA using a national database. Data from 424,354 patients from the ACS-NSQIP database was used to develop and validate four ML models to predict 30-day readmission risk after primary TKA. Individual model performance was assessed and compared based on discrimination, accuracy, calibration, and clinical utility. Length of stay (> 2.5days), body mass index (BMI) (> 33.21kg/m2), and operation time (> 93min) were important determinants of 30-day readmission. All ML models demonstrated equally good accuracy, calibration, and discriminatory ability (Brier score, ANN = RF = HGB = NEPLR = 0.03; ANN, slope = 0.90, intercept = - 0.11; RF, slope = 0.93, intercept = - 0.12; HGB, slope = 0.90, intercept = - 0.12; NEPLR, slope = 0.77, intercept = 0.01; AUCANN = AUCRF = AUCHGB = AUCNEPLR = 0.78). This study validates the generalizability of four previously developed ML algorithms in predicting readmission risk in patients undergoing TKA and offers surgeons an opportunity to reduce readmissions by optimizing discharge planning, BMI, and surgical efficiency.

Abstract 3547: The ISB Cancer Gateway in the Cloud (ISB-CGC): Access, explore and analyze large-scale cancer data through the Google Cloud

Abstract Rapid growth of cancer data in recent decades has made data discovery and wrangling difficult for the average cancer research lab. Our mission at the ISB Cancer Gateway in the Cloud (ISB-CGC), part of the NCI’s Cancer Research Data Commons ecosystem, is to democratize access to large cancer datasets. Funded by the NCI, we have performed ETL processes on data from GDC and PDC projects such as TCGA, TARGET, and CPTAC. We generated hundreds of BigQuery tables containing data such as mutations, gene expression, and protein abundance, which enable data analysis in the cloud via SQL. BigQuery analyses are inexpensive and rapid even when scaled to petabyte sized inputs, for example we ran 6.6 billion correlations in 2.5 hours with a total cost of about one dollar. These data can also be accessed affordably from Google Cloud VMs where researchers can develop analysis pipelines in Python, R, and workflow languages such as CWL. We present two recent collaborations: In one BigQuery was used to develop machine learning algorithms that calculated genetic risk scores from TCGA glioblastoma and ovarian cancer copy number variation. In another example researchers combined SQL queries of our BQ tables with data from the ISPY2 Trial initiative and generated an R shiny app that can dynamically create data visualizations for genes of interest in different TCGA cohorts. Citation Format: Fabian Seidl, Lauren Hagen, Jacob Wilson, Boris Aguilar, Deena Bleich, Lauren Wolfe, Poojitha Gundluru, Prema Venkatesan, Mi Tian, Suzanne Paquette, Elaine Lee, Danna Huffman, David Pot, William Longabaugh. The ISB Cancer Gateway in the Cloud (ISB-CGC): Access, explore and analyze large-scale cancer data through the Google Cloud [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 3547.

Abstract 3493: Read-level methylation pattern extraction for high-multiplex large-scale targeted NGS assay

Abstract The abnormal changes in DNA methylation are linked to the early stages of carcinogenesis. Identifying these epigenetic changes in circulating tumor DNA (ctDNA) can reveal potential biomarkers for the early diagnosis of various cancers. However, analyzing such data poses bioinformatics challenges due to the lack of sensitivity in detecting the low abundance ctDNA signals in biopsy samples, which are often overwhelmed by the complexity of libraries containing hundreds of targeted regions. Read-level methylation analysis holds the promise of more in-depth DNA methylation detection due to the wide coverage and high sensitivity of rare signals. However, this approach is hindered by the absence of a standardized workflow capable of generating interpretable reports suitable for both bench scientists and professional bioinformaticians. Here, we present a bioinformatics workflow that examines next-generation sequencing (NGS) data and characterizes the read-level methylation patterns of amplicons. Compared to other currently available tools, our method is designed to work with high-multiplex, large-scale targeted assays. It effectively eliminates the undesired noise derived from sequencing byproducts such as false CpG calls, dimers, and off-target alignments. Additionally, to accommodate the substantial volume of data generated by state-of-the-art NGS platforms, the workflow enables parallel processing of samples compatible with both cloud-based and on-premises computing resources. This workflow provides a comprehensive per-sample visualization of DNA methylation patterns and reports read-level methylation results in a “pattern-as-a-feature” table. In this table, the occurrence of an amplicon epiallelic haplotype (pattern) for every sample is represented as a “feature column” and is aggregated with all patterns discovered in the experiment. These read-level patterns, along with other information, can be used to develop machine learning algorithms to reiteratively harvest true predictive features and penalize confounding signals in predicting cancer diagnosis. Citation Format: Mingda Jin, Masatomo Kaneko, Steven Cen, Hongtao Li, Wei Guo, Xinyi Zhou, Atsuko Fujihara, Tsuyoshi Iwata, Lorenzo Storino Ramacciotti, Divyangi Paralkar, Giovanni E Cacciamani, Manju Aron, Osamu Ukimura, Inderbir S. Gill, Gangning Liang, Andre L. Abreu, Jeffrey Bhasin, Xiaojing Yang, Xi-Yu Jia. Read-level methylation pattern extraction for high-multiplex large-scale targeted NGS assay [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 3493.

Pulse shape discrimination in an organic scintillation phoswich detector using machine learning techniques

We developed machine learning algorithms for distinguishing scintillation signals from a plastic-liquid coupled detector known as a phoswich. The challenge lies in discriminating signals from organic scintillators with similar shapes and short decay times. Using a single-readout phoswich detector, we successfully identified γ radiation signals from two scintillating components. Our Boosted Decision Tree algorithm demonstrated a maximum discrimination power of 3.02 ± 0.85 standard deviation in the 950 keV region, providing an efficient solution for self-shielding and enhancing radiation detection capabilities.

Diagnostics and Assessment of the Severity of Alzheimer’s Disease: Machine Learning Algorithms Based on Markers of Inflammation

Background: as the most common form of dementia, Alzheimer’s disease (AD) is characterized by cognitive deterioration and usually begins with loss of memory of recent events. It is important to search for biological, sensitive and affordable methods that could be used for early diagnostics of AD and determine the severity of the disease.Objective: to develop machine learning algorithms based on such inflammatory markers as the enzymatic activity of leukocyte elastase (LE) and the functional activity of the α1-proteinase inhibitor (α1-PI) for diagnosing and assessing the severity of AD.Patients and methods: the study included128 people aged 55 to 94 years (73.7 ± 7.9 years), of which 91 patients were diagnosed with Alzheimer’s disease and 37 apparently healthy people (control). The indicators of LE and α1-PI in blood plasma were used as classifying features for building models. The following algorithms were used to build a machine learning model: Optimal Valid Partition (OVP), logistic regression (LR), support vector machine (SVM), random forest (RF), gradient boosting (GB) and statistically weighted syndromes (WSWS). The predictive performance of the constructed classiers was evaluated by the overall accuracy (accuracy), sensitivity (sensitivity), specificity (specificity), F-measure and ROC-analysis.Results: the developed machine learning algorithms made it possible to reliably divide the general group of subjects (patients + conditionally healthy), as well as patients with different AD severity, into 4 quadrants of a two-dimensional diagram in the LE and α1-PI coordinates and showed close and fairly high predictive efficiency.Conclusion: the developed machine learning algorithms have proven close and sufficiently high prognostic efficacy for assessing the severity of AD based on inflammatory markers (enzymatic activity of LE and functional activity of α1-PI) and, probably, can be useful for early diagnostics of the disease and timely administration of therapy.

Prediction of intraoperative red blood cell transfusion in valve replacement surgery: machine learning algorithm development based on non-anemic cohort.

Our study aimed to develop machine learning algorithms capable of predicting red blood cell (RBC) transfusion during valve replacement surgery based on a preoperative dataset of the non-anemic cohort. A total of 423 patients who underwent valvular replacement surgery from January 2015 to December 2020 were enrolled. A comprehensive database that incorporated demographic characteristics, clinical conditions, and results of preoperative biochemistry tests was used for establishing the models. A range of machine learning algorithms were employed, including decision tree, random forest, extreme gradient boosting (XGBoost), categorical boosting (CatBoost), support vector classifier and logistic regression (LR). Subsequently, the area under the receiver operating characteristic curve (AUC), accuracy, recall, precision, and F1 score were used to determine the predictive capability of the algorithms. Furthermore, we utilized SHapley Additive exPlanation (SHAP) values to explain the optimal prediction model. The enrolled patients were randomly divided into training set and testing set according to the 8:2 ratio. There were 16 important features identified by Sequential Backward Selection for model establishment. The top 5 most influential features in the RF importance matrix plot were hematocrit, hemoglobin, ALT, fibrinogen, and ferritin. The optimal prediction model was CatBoost algorithm, exhibiting the highest AUC (0.752, 95% CI: 0.662-0.780), which also got relatively high F1 score (0.695). The CatBoost algorithm also showed superior performance over the LR model with the AUC (0.666, 95% CI: 0.534-0.697). The SHAP summary plot and the SHAP dependence plot were used to visually illustrate the positive or negative effects of the selected features attributed to the CatBoost model. This study established a series of prediction models to enhance risk assessment of intraoperative RBC transfusion during valve replacement in no-anemic patients. The identified important predictors may provide effective preoperative interventions.

Experimental Investigations on the Thermal Characteristics of Domestic Convectors

Better understanding of local thermal characteristics of domestic convectors could play a crucial role in reducing energy consumption for space heating and decarbonizing the economy. The current study evaluates the impact of varying water inlet temperature and flowrate on the local surface temperature of domestic convectors through extensive empirical investigations. Experiments are performed using a custom-made test-rig featuring a 400 mm × 600 mm Type 11 convector within a large and well-ventilated environment, minimizing the thermal influence of the surrounding space on the thermal behavior of the convector. Infrared thermography (IR) is used to acquire local surface temperature data for further analysis. Based on the results obtained, it has been observed that the inlet water temperature has a negligible effect on thermal characteristics of the convector while increasing the flowrate substantially decreases the time required for the convector to reach maximum surface temperature. Based on the numerical data, an analytical model for average surface temperature has been developed using multiple variable regression analysis, demonstrating a prediction accuracy of >90% compared with the experimental data. A detailed understanding of the heating behavior exhibited by domestic convectors has led to a better understanding of the local thermal characteristics, while the prediction model can be used to develop machine learning algorithms to install better flow control techniques for efficient space heating.

AI-assisted image analysis and physiological validation for progressive drought detection in a diverse panel of Gossypium hirsutum L.

Drought detection, spanning from early stress to severe conditions, plays a crucial role in maintaining productivity, facilitating recovery, and preventing plant mortality. While handheld thermal cameras have been widely employed to track changes in leaf water content and stomatal conductance, research on thermal image classification remains limited due mainly to low resolution and blurry images produced by handheld cameras. In this study, we introduce a computer vision pipeline to enhance the significance of leaf-level thermal images across 27 distinct cotton genotypes cultivated in a greenhouse under progressive drought conditions. Our approach involved employing a customized software pipeline to process raw thermal images, generating leaf masks, and extracting a range of statistically relevant thermal features (e.g., min and max temperature, median value, quartiles, etc.). These features were then utilized to develop machine learning algorithms capable of assessing leaf hydration status and distinguishing between well-watered (WW) and dry-down (DD) conditions. Two different classifiers were trained to predict the plant treatment-random forest and multilayer perceptron neural networks-finding 75% and 78% accuracy in the treatment prediction, respectively. Furthermore, we evaluated the predicted versus true labels based on classic physiological indicators of drought in plants, including volumetric soil water content, leaf water potential, and chlorophyll a fluorescence, to provide more insights and possible explanations about the classification outputs. Interestingly, mislabeled leaves mostly exhibited notable responses in fluorescence, water uptake from the soil, and/or leaf hydration status. Our findings emphasize the potential of AI-assisted thermal image analysis in enhancing the informative value of common heterogeneous datasets for drought detection. This application suggests widening the experimental settings to be used with deep learning models, designing future investigations into the genotypic variation in plant drought response and potential optimization of water management in agricultural settings.

PATIENT-SPECIFIC ADAPTIVE DYNAMIC CYCLING IS ASSOCIATED WITH IMPROVED PD SYMPTOMS

BACKGROUND High-cadence dynamic cycling is an effective therapy for improving motor symptoms in individuals with Parkinson’s Disease (PD), as measured by the Unified Disease Parkinson’s Rating Scale-Motor III (UPDRS-III). However, there is significant variability in individual responses to this therapy. Our lab developed a patient-specific adaptive dynamic cycling (PSADC) paradigm that manipulates entropy of cadence to optimize exercise prescriptions for individuals at various stages of disease progression. The purpose of this study is to evaluate the effectiveness of 12 sessions of PSADC on motor symptom improvement (UPDRS-III score) in individuals with PD. METHODS Twenty-three individuals with idiopathic PD (were randomized into two groups: PSADC (n=13) or active control (n=10). All individuals completed 12 sessions (3 sessions per week for 4 weeks) of dynamic cycling on a SMART (Speed Manipulated Adaptive Rehabilitation Therapy) bicycle. Each session consisted of a 5-minute warm-up at 60 revolutions per minute (rpm), 30-minute exercise session (80 rpm), and 5-minute cool-down (60 rpm). Individuals in the PSADC group followed an adaptive exercise prescription in which resistance level was optimized on a weekly basis, based on the individual’s entropy of cadence, and cycling effort. Individuals in the active control group remained at a constant resistance level for the entirety of the intervention. UPDRS-III was assessed in all participants prior to and following the 12-session intervention. Two-way analysis of variance (ANOVA) and paired samples t-tests were performed to detect statistical differences in UPDRS-III score between the groups. RESULTS There was a significant group by time interaction (F= 18.746, p <0.001). The PSADC group showed a significant reduction (improvement) in UPDRS-III score (Pre: 32.8, Post: 27.5; p<0.001), while the active control group showed no significant change in UPDRS-III score (Pre: 28.2, Post: 32.4, p=0.08). CONCLUSIONS 12 sessions of PSADC significantly improved UPDRS-III score, compared to non-adaptive high-cadence dynamic cycling. These results suggest that optimizing entropy of cadence is valuable for motor symptom improvement. Future studies will develop machine learning algorithms designed to predict appropriate clinical exercise prescriptions for individuals with PD.

Use of novel cancer staging model to analyze healthcare costs of commercially insured patients with colorectal cancer.

35 Background: Earlier colorectal cancer diagnosis can improve patient outcomes, but the payer cost impact of earlier stage at cancer diagnosis is not well understood because stage information is not directly available in administrative claims data. This data gap impairs researchers’ ability to fully evaluate the financial impact of earlier neoplasia detection resulting from colorectal cancer screening. In prior work, we used SEER-Medicare data to develop machine learning algorithms to assign AJCC stages to colon (CC) and rectal (RC) cancer patients using treatment flags derived from claims data. These algorithms were 83% (CC) and 69% (RC) accurate at assigning incident cancer patients to their SEER-recorded AJCC stage. This new work applied those validated algorithms to commercial claims data to study the healthcare costs by stage of patients identified with incident CC or RC. Methods: Newly diagnosed CC and RC patients were identified in Milliman’s 2019-2020 Consolidated Health Cost Guidelines Sources Data for commercially insured patients. Patients were assigned a cancer stage using a claims-based predictive multinomial logistic regression model (R Statistical Software - v4.1.2 R Core Team 2021; nnet package - Venables and Ripley 2002). Allowed medical and pharmacy costs were summarized by month, starting from each patient’s initial claim with a cancer diagnosis through the end of the patient’s enrollment or through September 2022. Patients with each cancer type were then grouped by stage, and mean cumulative costs at 12-, 24-, and 36-months were calculated. Results: CC patients with stage 4 disease incur costs over 4x greater than those with stages 2B or lower and 1.7x greater than patients with stages 2B/3 during the first year after diagnosis. Cost differentials by stage are smaller among patients with RC. RC patients with stage 4 disease incur costs 2.2x greater than those with stages 2B or lower and 1.3x greater than patients with stages 2B/3 over the first year after diagnosis. For both CC and RC, substantial differences in cost persist over a 3-year period post-diagnosis. Conclusions: We applied a novel cancer staging model to commercial claims to estimate patients’ healthcare costs by stage at diagnosis. CC and RC patients’ costs increase substantially as their stages at diagnosis increase. In addition to improving patient outcomes, diagnosis of CC and RC at earlier stages among commercial patients could reduce treatment costs. [Table: see text]

Machine learning models using symptoms and clinical variables to predict coronary artery disease on coronary angiography.

Coronary angiography (CAG) is invasive and expensive, while numbers of patients suspected of coronary artery disease (CAD) undergoing CAG results have no coronary lesions. To develop machine learning algorithms using symptoms and clinical variables to predict CAD. This study was conducted as a cross-sectional study of patients undergoing CAG. We randomly chose 2082 patients from 2602 patients suspected of CAD as the training set, and 520 patients as the test set. We utilized LASSO regression to do feature selection. The area under the receiver operating characteristic curve (AUC), confusion matrix of different thresholds, positive predictive value (PPV) and negative predictive value (NPV) were shown. Support vector machine algorithm performances in 10 folds were conducted in the training set for detecting severe CAD, while XGBoost algorithm performances were conducted in the test set for detecting severe CAD. The algorithm of logistic regression achieved an average AUC of 0.77 in the training set during 10-fold validation and an AUC of 0.75 in the test set. When probability predicted by the model was less than 0.1, 11 patients in the test set (520 patients) were screened out, and NPV reached 90.9%. When probability predicted by the model was less than 0.2, 110 patients in the test set were screened out, and reached 83.6%. Meanwhile, when threshold was set to 0.9, PPV reached 97.4%. When the threshold was set to 0.8, PPV reached 91.5%. Machine learning algorithm using data from hospital information systems could assist in severe CAD exclusion and confirmation, and thus help patients avoid unnecessary CAG.

Machine learning assessment of dental age classification based on cone-beam CT images: a different approach.

Machine learning (ML) algorithms are a portion of artificial intelligence that may be used to create more accurate algorithmic procedures for estimating an individual's dental age or defining an age classification. This study aims to use ML algorithms to evaluate the efficacy of pulp/tooth area ratio (PTR) in cone-beam CT (CBCT) images to predict dental age classification in adults. CBCT images of 236 Turkish individuals (121 males and 115 females) from 18 to 70 years of age were included. PTRs were calculated for six teeth in each individual, and a total of 1416 PTRs encompassed the study dataset. Support vector machine, classification and regression tree, and random forest (RF) models for dental age classification were employed. The accuracy of these techniques was compared. To facilitate this evaluation process, the available data were partitioned into training and test datasets, maintaining a proportion of 70% for training and 30% for testing across the spectrum of ML algorithms employed. The correct classification performances of the trained models were evaluated. The models' performances were found to be low. The models' highest accuracy and confidence intervals were found to belong to the RF algorithm. According to our results, models were found to be low in performance but were considered as a different approach. We suggest examining the different parameters derived from different measuring techniques in the data obtained from CBCT images in order to develop ML algorithms for age classification in forensic situations.

Experimental Analysis on Performance of Speech Utterance recognition using AI Models

In Automated speech recognition of the system performance is crucial and important to satisfy multiple requirements of HMI and, more recently, even in IoT-related applications as well. Concurrently, there has been an increase in demand for detecting strong critical features derived from speech utterances. This paper presents a performance of the developed machine learning algorithms with respect to audio digit speech recognition and classification. The prepared dataset contains a free range of words (from 1 to 10) from speakers of different age groups. The Audacity software used for preprocessing the audio files that includes removal of noise included in the signal and trimming the silence on either side of the word utterance. audio signal sampled at fs = 48kHz.We have developed four AI Models to recognise the word utterances. Audio signals are processed separately and derived two unique feature sets that includes statistical features set and singular values by performing SVD related to word utterances. The cepstral values for each utterance are obtained from state-of-the-art MFCC. Variance-covariance matrix is calculated from the generated MFCC matrix. The diagonal values which form the variance are recorded and denoted as feature set-1 for the word utterance and inputted to the machine learning algorithms. Performance matrices of the developed models are recorded. To keep the computational bottleneck associated with the use of feature sets to minimum, dimensionality reduction is carried out by applying singular value decomposition to the extracted MFCC matrix. The derived set of singular values ​​considered as feature set-2 is used to train and test the developed AI models with a ratio of 70:30. We presented and discussed the performance and results produced by MLP, KNN, SVM, Random Forest algorithms. In comparison, MLP and Random Forest were found to show excellent performance on both feature sets with 100% training accuracy and 99% test accuracy.

Machine Learning Validates Risk Biomarkers of Chronic Graft-Versus-Host Disease in 936 Patients from BMT CTN 0201 & 1202 Cohorts

Introduction Our previous work identified susceptibility/risk biomarkers for developing chronic GVHD (cGVHD) in individuals without clinically apparent disease (Logan et al. 2023, JCI), which is a major contributor to morbidity and mortality following allogeneic hematopoietic cell transplant (HCT). In this study, we used machine learning (ML) algorithms to build biomarker-based prediction models for development of cGVHD and non-relapse mortality (NRM). Methods Data were obtained on 9 pre-transplant factors and 7 plasma proteins at Day 90 post HCT in 936 HCT recipients from the BMT CTN 0201 and 1202 studies. The study population was split into separate training (80%) and validation (20%) datasets. Associations of each protein marker with the hazards of cGVHD and NRM were evaluated using Cox proportional hazards (PH) models. To evaluate the predictive ability of the proteins, ML models were constructed from them and the time varying Area Under the ROC curve (AUCt) was estimated at Days 180, 270, 360, and 540 post HCT. We considered ML approaches within the PH model framework, including Boosting (XGBoost), Group SCAD, and Adaptive Group Lasso, as well as methods avoiding the PH assumption, including Random Survival Forests and Bayesian Additive Regression Trees (BART). DeepSurv and DeepHit models were also used in the first application of deep learning to GVHD biomarker risk evaluation to our knowledge. A p-value cutoff of 0.05 determined statistical significance. Results Of the 7 proteins tested, 5 (CXCL9, CXCL10, MMP3, DKK3, and CD163) were associated with cGVHD risk, while 4 (MMP3, DKK3, ST2, CD163) were correlated with NRM (p<0.05 for all). Moreover, effects of some markers varied depending on the graft type, with higher MMP3 found to have a larger, deleterious effect on cGVHD risk in patients with bone marrow (BM) compared to peripheral blood (PB) graft (p=0.03). CD163 had a similar interaction with graft type such that its effects on cGVHD and NRM risk are significant with BM graft but not with PB graft (p<0.05 for both). The abilities of the ML methods to risk stratify cGVHD and NRM were assessed and contrasted to a Group SCAD model having clinical factors only. Because graft type was found to impact both the risks of cGVHD and NRM as well as the biomarkers' effects, the ML models considered markers, graft type, and their interactions. In modeling cGVHD risk, the Group SCAD model with only clinical factors selected 4 variables: age, graft type, donor-recipient sex-mismatch, and GVHD prophylaxis; while Adaptive Group Lasso and Group SCAD with biomarkers selected 5 proteins: CXCL9, CXCL10, MMP3, ST2, and CD163. In the validation dataset, all ML methods with biomarkers provided similar or higher AUCt for cGVHD after Day 270 compared to clinical factors only, suggesting that plasma proteins measured as early as Day 90 can inform underlying cGVHD biology that has not manifested clinically ( Fig). BART and Boosting yielded the highest AUCt and were better than the model with only clinical factors, attaining AUCt > 0.60 from Day 270-540. To assess the variables' importance in ML, we examined BART's posterior probabilities of selection, which describe how frequently each variable is used in the BART tree ensemble. For cGVHD, variables with selection probabilities > 0.10 were graft type, CXCL9, and MMP3. Notably, the protein markers that were significant in separate models also had the largest selection probabilities of all markers. The AUCt for NRM was stable across time points in the 0.65-0.72 range and superior to clinical factors after Day 180 ( Fig). For NRM, variables with selection probabilities > 0.10 were graft type, ST2, MMP3, CD163, and DKK3. Conclusions ML methods using non-invasive plasma proteins can successfully identify and validate risk biomarkers of cGVHD. ML algorithms using objective and early measurements of soluble markers including CXCL9, CXCL10, MMP3, DKK3, and ST2 perform better in predicting cGVHD than ML with known clinical factors only. ML with MMP3, DKK3, ST2, and CD163 improved NRM prediction. Deep learning models did not outperform more classical ML methods, possibly due to a limited sample size that could not reveal intricate relationships between proteins and cGVHD/NRM risk. Several proteins represent potential therapeutic targets. These data support future research to further validate these biomarkers and to develop ML algorithms to identify patients at risk for developing cGVHD and NRM.

Machine Learning Algorithms Predict Successful Weaning From Mechanical Ventilation Before Intubation: Retrospective Analysis From the Medical Information Mart for Intensive Care IV Database.

The prediction of successful weaning from mechanical ventilation (MV) in advance of intubation can facilitate discussions regarding end-of-life care before unnecessary intubation. We aimed to develop a machine learning-based model that predicts successful weaning from ventilator support based on routine clinical and laboratory data taken before or immediately after intubation. We used the Medical Information Mart for Intensive Care IV database, which is an open-access database covering 524,740 admissions of 382,278 patients in Beth Israel Deaconess Medical Center, United States, from 2008 to 2019. We selected adult patients who underwent MV in the intensive care unit (ICU). Clinical and laboratory variables that are considered relevant to the prognosis of the patient in the ICU were selected. Data collected before or within 24 hours of intubation were used to develop machine learning models that predict the probability of successful weaning within 14 days of ventilator support. Developed models were integrated into an ensemble model. Performance metrics were calculated by 5-fold cross-validation for each model, and a permutation feature importance and Shapley additive explanations analysis was conducted to better understand the impacts of individual variables on outcome prediction. Of the 23,242 patients, 19,025 (81.9%) patients were successfully weaned from MV within 14 days. Using the preselected 46 clinical and laboratory variables, the area under the receiver operating characteristic curve of CatBoost classifier, random forest classifier, and regularized logistic regression classifier models were 0.860 (95% CI 0.852-0.868), 0.855 (95% CI 0.848-0.863), and 0.823 (95% CI 0.813-0.832), respectively. Using the ensemble voting classifier using the 3 models above, the final model revealed the area under the receiver operating characteristic curve of 0.861 (95% CI 0.853-0.869), which was significantly better than that of Simplified Acute Physiology Score II (0.749, 95% CI 0.742-0.756) and Sequential Organ Failure Assessment (0.588, 95% CI 0.566-0.609). The top features included lactate and anion gap. The model's performance achieved a plateau with approximately the top 21 variables. We developed machine learning algorithms that can predict successful weaning from MV in advance to intubation in the ICU. Our models can aid the appropriate management for patients who hesitate to decide on ventilator support or meaningless end-of-life care.

Evidence-based machine learning algorithm to predict failure following cartilage procedures in the knee

IntroductionClinical decision-making is highly based on expert opinion. Machine learning is increasingly used to develop patient-specific risk prediction analysis to improve patient selection prior to surgery. ObjectivesTo develop machine learning algorithms to predict failure of surgical procedures that address cartilage defects of the knee and detect variables associated with failure. MethodsAn institutional database was queried for cartilage procedures performed between 2000 and 2018. Failure was defined as revision cartilage surgery or knee arthroplasty. One hundred and one preoperative and intraoperative features were evaluated as potential predictors. Four machine learning algorithms were trained and internally validated. ResultsOne thousand and ninety-one patients with a minimum follow-up of 2 years were included and underwent chondroplasty (n = 560; 51%), osteochondral allograft transplantation (n = 306; 28%), microfracture (n = 150; 14%), autologous chondrocyte implantation (n = 39; 4%), or osteochondral autograft transplantation (n = 36; 3%). The Random Forest algorithm was the best-performing algorithm, with an area under the curve of 0.765 and a Brier score of 0.135. The most important features for predicting failure were symptom duration, age, body mass index, lesion grade, and total lesion area. Local Interpretable Model-agnostic Explanations analysis provided patient-specific comparisons for the risk of failure of an individual patient being assigned various types of cartilage procedures. ConclusionsMachine learning algorithms were accurate in predicting the risk of failure following cartilage procedures of the knee, with the most important features in descending order being symptom duration, age, body mass index, lesion grade, and total lesion area. Machine learning algorithms may be used to compare the risk of failure of specific patient-procedure combinations in the treatment of cartilage defects of the knee.

Critical Assessment of pH-Dependent Lipophilicity Profiles of Small Molecules: Which One Should We Use and In Which Cases?

Lipophilicity is a physicochemical property with wide relevance in drug design, computational biology, and food, environmental and medicinal chemistry. Lipophilicity is commonly expressed as thepartition coefficientfor neutral molecules, whereas for molecules with ionizable groups, the distribution coefficient (D) at a given pH is used. The logDpH is usually predicted using a pH correction, while often ignoring the apparent ionic partition [[EQUATION]]. In this work, we studied the impact of [[EQUATION]] on the prediction of both the experimental lipophilicity of small molecules and experimental lipophilicity-based applications and metrics such as lipophilic efficiency (LipE), distribution of spiked drugs in milk products, and pH-dependent partition of water contaminants in synthetic passive samples such as silicones. Our findings show that better predictions are obtained by considering the apparent ionic partition.In this context, we developed machine learning algorithms to determine the cases that [[EQUATION]] should be considered. The results indicate that small, rigid, and unsaturated molecules with logPN close to zero, which present a significant proportion of ionic species in the aqueous phase, were better modeled using the apparent ionic partition [[EQUATION]]). Finally, our findings can serve as guidance to the scientific community working in early-stage drug design, food, and environmental chemistry.