SEARCH · PubMed Health

Results for “machine learning explanation”

Explore indexed PubMed citations for clinical trials, systematic reviews and public health research. Read source abstracts and follow each citation to its original PubMed record.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

47 records · Page 3Linked to original sources

PubMed1996

Machine learning in prognosis of the femoral neck fracture recovery.

We compare the performance of several machine learning algorithms in the problem of prognostics of the femoral neck fracture recovery: the K-nearest neighbours algorithm, the semi-naive Bayesian classifier, backpropagation with weight elimination learning of the multilayered neural networks, the LFC (lookahead feature construction) algorithm, and the Assistant-I and Assistant-R algorithms for top down induction of decision trees using information gain and RELIEFF as search heuristics, respectively. We compare the prognostic accuracy and the explanation ability of different classifiers. Among the different algorithms the semi-naive Bayesian classifier and Assistant-R seem to be the most appropriate. We analyze the combination of decisions of several classifiers for solving prediction problems and show that the combined classifier improves both performance and the explanation ability.

Algorithms↗

PubMed2026

Development and external validation of an explainable machine learning model for predicting chronic kidney disease progression in the Korean population.

BACKGROUND: Current risk stratification models, such as the Kidney Failure Risk Equation (KFRE), exhibit variable performance across ethnic groups and fail to capture dynamic clinical trajectories. This study aimed to develop and validate a Korean-specific machine learning (ML) model for predicting chronic kidney disease (CKD) progression using an ensemble approach. METHODS: We used electronic health records from Seoul National University Hospital for model development (n = 28,209) and the Korean Genome and Epidemiology Study (KoGES) CKD cohort for external validation (n = 3,960). The primary outcome was a composite of ≥40% decline in estimated glomerular filtration rate (eGFR) or progression to end-stage renal disease within 2 years. A soft-voting ensemble of four ML algorithms (XGBoost, LightGBM, CatBoost, and Random Forest) was developed. RESULTS: The ensemble model demonstrated robust discrimination in internal validation (area under the receiver operating characteristic curve [AUROC], 0.939; 95% confidence interval [CI], 0.934-0.944), significantly exceeding the KFRE (AUROC, 0.879-0.884). External validation in the KoGES cohort showed comparable discrimination (AUROC, 0.859; 95% CI, 0.798-0.914) versus KFRE (four-variable AUROC, 0.882; 95% CI, 0.818-0.935). Shapley Additive exPlanations (SHAP) analysis identified baseline eGFR, serum creatinine, eGFR slope, albumin, and hemoglobin as key prognostic features, supporting a complementary framework using KFRE for community screening and the ML model for hospital-based risk stratification. CONCLUSION: The ensemble ML model accurately predicts short-term CKD progression in Korean patients. By incorporating longitudinal features and ensemble learning, it provides a precise alternative to Western-derived equations, particularly in tertiary care settings.

Chronic kidney failure↗

PubMed1975

Escalation of aggression: experimental studies.

A finding commonly obtained in research using the Buss "aggression machine" is a main effect for trail blocks, indicating an escalation in shock intensity over trails. Theoretical explanations for this effect were tested in a modified verbal operant-conditioning situation. In Experiment 1, subjects could administer any of 10 levels of positive reinforcement to a "learner" for correct verbal responses or any of 10 levels of negative reinforcement to a learner for incorrect responses. Half of the subjects were required to begin with weak, half with strong, reinforcements. Results indicated that, regardless of condition, subjects gave more intense reinforcements as the learning trails progressed. Those who administered negative reinforcements devalued the learner relative to those who administered positive reinforcements. In Experiment 2, a role-playing procedure was used in which subjects administered either positive or negative reinforcements to a learner whose performance either did or did not improve over trials. Again, in all experimental groups, subjects administered increasingly intense reinforcements over trials. The results are interpreted as supporting a disinhibition theory of anti- and prosocial behavior.

Aggression↗

PubMed1997

Induction of decision trees and Bayesian classification applied to diagnosis of sport injuries.

Machine learning techniques can be used to extract knowledge from data stored in medical databases. In our application, various machine learning algorithms were used to extract diagnostic knowledge which may be used to support the diagnosis of sport injuries. The applied methods include variants of the Assistant algorithm for top-down induction of decision trees, and variants of the Bayesian classifier. The available dataset was insufficient for reliable diagnosis of all sport injuries considered by the system. Consequently, expert-defined diagnostic rules were added and used as pre-classifiers or as generators of additional training instances for diagnoses for which only few training examples were available. Experimental results show that the classification accuracy and the explanation capability of the naive Bayesian classifier with the fuzzy discretization of numerical attributes were superior to other methods and estimated as the most appropriate for practical use.

Artificial Intelligence↗

PubMed2000

Radical pruning: a method to construct skeleton radial basis function networks.

Trained radial basis function networks are well-suited for use in extracting rules and explanations because they contain a set of locally tuned units. However, for rule extraction to be useful, these networks must first be pruned to eliminate unnecessary weights. The pruning algorithm cannot search the network exhaustively because of the computational effort involved. It is shown that using multiple pruning methods with smart ordering of the pruning candidates, the number of weights in a radial basis function network can be reduced to a small fraction of the original number. The complexity of the pruning algorithm is quadratic (instead of exponential) in the number of network weights. Pruning performance is shown using a variety of benchmark problems from the University of California, Irvine machine learning database.

Algorithms↗

PubMed2006

Personality traits in miners with past occupational elemental mercury exposure.

In this study, we evaluated the impact of long-term occupational exposure to elemental mercury vapor (Hg0) on the personality traits of ex-mercury miners. Study groups included 53 ex-miners previously exposed to Hg0 and 53 age-matched controls. Miners and controls completed the self-reporting Eysenck Personality Questionnaire and the Emotional States Questionnaire. The relationship between the indices of past occupational exposure and the observed personality traits was evaluated using Pearson's correlation coefficient and on a subgroup level by machine learning methods (regression trees). The ex-mercury miners were intermittently exposed to Hg0 for a period of 7-31 years. The means of exposure-cycle urine mercury (U-Hg) concentrations ranged from 20 to 120 microg/L. The results obtained indicate that ex-miners tend to be more introverted and sincere, more depressive, more rigid in expressing their emotions and are likely to have more negative self-concepts than controls, but no correlations were found with the indices of past occupational exposure. Despite certain limitations, results obtained by the regression tree suggest that higher alcohol consumption per se and long-term intermittent, moderate exposure to Hg0 (exposure cycle mean U-Hg concentrations > 38.7 < 53.5 microg/L) in interaction with alcohol remain a plausible explanation for the depression associated with negative self-concept found in subgroups of ex-mercury miners. This could be one of the reason for the higher risk of suicide among miners of the Idrija Mercury Mine in the last 45 years.

Adult↗

PubMed2026

From prediction to mechanism: Explainable AI uncovers plasma and CSF proteomic signatures of Alzheimer's disease.

Alzheimer's disease (AD) plasma and cerebrospinal fluid (CSF) proteomics can distinguish AD from cognitively normal controls, but the generalizability of machine learning performance and the recurrence of biological signals across datasets require cautious interpretation. We developed an explainable artificial intelligence framework spanning two fluids and four ADNI proteomic datasets, covering 2082 modality specific samples, all analysed internally within ADNI. Phase 1 analysed plasma using a 119 analyte NULISA and targeted UPENN panel (n = 727; 216 CE, 511 controls). Phase 2 extended the analysis to CSF using SOMAscan7k, TMT-MS and targeted SET2, with Elecsys Aβ42, Aβ40, total tau and p-tau181 as anchor biomarkers. Only SOMAscan was subject-independent relative to Phase 1 plasma; TMT-MS and SET2 overlapped with Phase 1 for 96.0% and 97.7% of subjects and therefore are not independent replication cohorts. Under subject-level splits with fold internal preprocessing, we compared Elastic Net, Explainable Boosting Machines and gradient boosted trees with SHAP-based explanations. Among the candidate pipelines, we selected the pipeline with the highest held-out test ROC AUC for each platform; the selected values were 0.927 in plasma and 0.954-0.973 across the three CSF datasets. Because the same held out test performance was used for pipeline selection and headline reporting, these are optimistically selected single-holdout estimates, not unbiased estimates of generalizable or clinical performance. Explanations identified five recurring biological axes within ADNI: cholinergic (ACHE), tau/14-3-3 (YWHAG, YWHAZ, YWHAB, YWHAE), neuro-axonal (NEFL, NEFH), microglial/complement (CHIT1, SMOC1, CHI3L1, C7, CFH) and synaptic (NPTXR, NPTX2, DLG4, SYT5, VSNL1, ELAVL2). CSF analyses showed synaptic vesicle-cycle enrichment (q = 2 × 10-6), and CSF YWHAG correlated strongly with total tau (ρ = 0.87). Cross-fluid directional concordance was modest overall (54-57%) but increased to 73-80% among mapped analyte/protein rows reaching q < 0.05 in CSF. These findings provide hypothesis-generating, internally supported evidence within ADNI. Independent external cohorts with locked pipelines are required to evaluate generalizable performance and biological reproducibility; the overlapping TMT-MS and SET2 analyses should not be interpreted as independent replication.

Alzheimer Disease↗

PubMed2003

Explaining the output of ensembles in medical decision support on a case by case basis.

The use of ensembles in machine learning (ML) has had a considerable impact in increasing the accuracy and stability of predictors. This increase in accuracy has come at the cost of comprehensibility as, by definition, an ensemble model is considerably more complex than its component models. This is of significance for decision support systems in medicine because of the reluctance to use models that are essentially black boxes. Work on making ensembles comprehensible has so far focused on global models that mirror the behaviour of the ensemble as closely as possible. With such global models there is a clear tradeoff between comprehensibility and fidelity. In this paper, we pursue another tack, looking at local comprehensibility where the output of the ensemble is explained on a case-by-case basis. We argue that this meets the requirements of medical decision support systems. The approach presented here identifies the ensemble members that best fit the case in question and presents the behaviour of these in explanation.

Anticoagulants↗

PubMed2026

Transcriptome-based high-frequency recurrence index predicts frequent recurrence in non-muscle-invasive bladder cancer after Bacillus Calmette-Guérin therapy.

BACKGROUND: High-frequency recurrence (HfR, ≥ 2 recurrences) in non-muscle-invasive bladder cancer (NMIBC) poses a significant clinical burden. Current risk models, such as the European Organization for Research and Treatment of Cancer (EORTC), the European Association of Urology (EAU), and the UROMOL classification, offer limited predictive accuracy for identifying patients at risk for frequent recurrence despite appropriate treatment. METHODS: A 75-gene high-frequency recurrence index (HfRI) was constructed by selecting recurrence-associated genes using differential expression and Cox regression analyses. The HfRI was computed as a weighted sum of normalized gene expression values. The model was trained on a discovery cohort and validated in multiple cohorts (n = 1379) using machine-learning approaches. Clinical relevance was assessed using recurrence-free survival (RFS) and Cox models, and predictive performance was compared with that of the EORTC, EAU, and UROMOL classifications using the area under the curve (AUC) and the concordance index (c-index). RESULTS: The HfRI robustly stratified patients into high-risk and low-risk groups across six independent NMIBC cohorts. Patients classified as HfRI-high had a significantly greater likelihood of experiencing ≥ 2 recurrences (χ2, p = 0.001) and showed markedly reduced RFS (log-rank test, p < 0.001). The adverse prognostic effect of the HfRI persisted even among patients treated with BCG therapy (log-rank test, p = 0.02). Multivariate analysis revealed that the HfRI was an independent predictor of HfR (HR = 2.82, 95% CI = 1.89-4.20, p < 0.001). Compared with established clinical risk classifiers, the HfRI demonstrated superior predictive performance (AUC = 0.736, c-index = 0.673) in terms of the EORTC (AUC = 0.594), EAU (AUC = 0.557) risk groups, and UROMOL2021 (AUC = 0.596) classification. Pathway analysis revealed that HfRI-high tumors were characterized by upregulation of cell cycle progression and DNA replication pathways, accompanied by suppression of immune signaling pathways. These biological features provide a mechanistic explanation for the reduced responsiveness to intravesical BCG therapy, underscoring the role of HfRI not only as a predictor of recurrence risk but also as a biomarker capable of identifying patients unlikely to benefit from standard BCG treatment. CONCLUSIONS: HfRI represents a robust, transcriptome-based tool for predicting frequent recurrence in NMIBC patients. The HfRI supports earlier identification of patients at risk of high-frequency recurrence, thereby supporting personalized treatment strategies.

Humans↗

PubMed2001

Hibernation effects on memory in European ground squirrels (Spermophilus citellus).

Effects of hibernation on memory were tested in European ground squirrels (Spermophilus citellus). The animals were trained in summer to successfully accomplish two tasks: a spatial memory task in a maze and an operant task on a feeding machine. One group hibernated normally, and the other was prevented from hibernation by maintaining ambient temperature at 22 degrees C. In spring, the same tasks were repeated for both groups and their individual performances compared to the initial training phase. The experimental groups differed significantly in both tasks. The nonhibernating animals had higher levels of retention and needed significantly fewer trials to relearn the tasks than the group that had hibernated. In addition to testing the retention of conditioned tasks, social memory was also studied. The ground squirrels were given a social recognition test in spring with one familiar and one unfamiliar conspecific. In contrast to the conditioned tasks, social memory did not seem to be affected by hibernation. The results indicate negative effects of hibernation on the retention of conditioned tasks, which could produce important constraints on animals. A potential explanation for this memory loss might be changes in neuronal activity, which occur during hibernation.

Animals↗

PubMed1995

An interactive consultation multimedia software for orthodontic patients.

Presentation of diagnosis and treatment planning for orthodontic problems by orthodontists is often a hurdle and a nuisance to most patients. The reasons are that it has much content which may be hard to understand without having expert knowledge related to the temporal change in dentofacial structures known as the growth, development and physiological aspects of masticatory apparatus. To complement this, we have developed an interactive consultation multimedia software for orthodontic patients. he design concept of the current software has three aspects. Firstly, since the software is operated by orthodontic patients themselves or by their parents, it enhances the operational feasibility. Secondly, it helps the patients choose the information in which they are interested. Thirdly, it emphasizes audio-visual understanding of orthodontic practice, including terminology. e used a hypertext machine with a 240MB hard disk drive, an 8MB RAM and a 13 inch color monitor. In developing the current software, we also used a video camera, a video color board, a microphone, and an image scanner together with an image recorder, a movie and sound data editing system, image scanning and editing, an image changer, a spread sheet and mathematical software. he current software consists of various multimedia such as images, sounds, characters, and biosignals. The "stack" of the software consists of three parts: a) "General Understanding of Occlusion" b) "Understanding Specific Types of Occlusion Exhibited by the Patient" c) "Orthodontic Terminology" When card A is selected the patient can choose either "Good Occlusion" or "Malocclusion." If "Malocclusion" is chosen, respective occlusal types are shown. The next card provides pathological conditions caused by respective malocclusion, e.g., gingivitis. After selecting card B which asks the patient, "What do your teeth look like?" the following buttons are provided: "Maxillary Protrusion," "Reversed Occlusion," "Crowding," "Open Bite," and "Spaced Arch." After selecting one of these, the card with an explanation of the respective malocclusion is shown according to the patient's physiological age. Finally, after card C is selected, a new card which has a list of orthodontic terminology is presented. Patients can search any term according to their choice to open a new card which gives a detailed explanation. e confirmed that the current consultation multimedia software can provide a comfortable environment to the patients and their families to learn where the orthodontic problems lie and how they could be solved.

Audiovisual Aids↗