Andreea Alexandra Anghel
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Andreea Alexandra Anghel has a total of 28 co-authors. Here are the top co-authors by number of shared publications:
- Catalin Anghel
Department of Computer Science and Information Technology, Faculty of Automation, Computers, Electrical Engineering and Electronics, “Dunarea de Jos” University of Galati, 800146 Galati, Romania
20 shared publications - Adina Cocu
Department of Computer Science and Information Technology, “Dunărea de Jos” University of Galati, Științei St. 2, 800146 Galati, Romania
18 shared publications - Marian Viorel Craciun
Department of Computer Science and Information Technology, “Dunărea de Jos” University of Galati, Științei St. 2, 800146 Galati, Romania
16 shared publications - Constantin Adrian Andrei
Faculty of Medicine, The “Carol Davila” University of Medicine and Pharmacy, 050474 Bucharest, Romania
10 shared publications - Cristian Scheau
The “Carol Davila” University of Medicine and Pharmacy, 050474 Bucharest, Romania
8 shared publications - Serban Dragosloveanu
The “Carol Davila” University of Medicine and Pharmacy, 050474 Bucharest, Romania
8 shared publications
Latest Publications
- Open Access
- 10 October 2026
- Published by MDPI AG in Machine Learning and Knowledge Extraction
- Vol. 8 (10) , 327
ClaimArena-LLM: Auditing Adversarial Propagation of False Causal Attributions in Multi-Agent LLM Debates
Background: Multi-agent large language model (LLM) systems increasingly rely on structured interaction, debate, and final model-based evaluation, yet an unsupported causal attribution introduced by one participant may propagate through the interaction and influence subsequent decisions. This study introduces ClaimArena-LLM, a controlled framework for auditing adversarial propagation, evidentiary provenance, and final-stage evaluator vulnerability in multi-agent causal reasoning. Methods: ClaimArena-LLM uses 100 synthetic causal incident cases with predefined evidence-supported and false causal propositions. Each case was evaluated under matched adversarial and all-honest debate conditions using four functional roles: investigator, honest analyst, adversarial analyst, and auditor. The experimental program comprised 1460 validated runs, including the 200-run main experiment, claim-source lineage visibility and debate-round ablations, an 800-run balanced model–role rotation, and a 60-run proposition-identifier and presentation-order sensitivity control. Outcomes were scored deterministically from structured causal propositions, evidence references, claim lineage, reports, and auditor verdicts. Results: The adversarial analyst maintained the predefined false causal target with 97.6% mean fidelity. Adversarial exposure increased investigator shifts toward the false target by 16.0 percentage points (Holm-adjusted p = 0.035). At the auditor stage, false-target attribution increased by 11.0 percentage points (adjusted p = 0.043), verdict contamination by 20.0 percentage points (adjusted p = 0.002), and semantic provenance validity decreased by 20.0 percentage points (adjusted p = 0.002). Neither the manipulation nor the false target was explicitly detected. The ablations showed no robust adjusted differences, model–role rotation revealed substantial configuration dependence, and no benchmark-characteristic interaction remained significant after Holm correction. The proposition-identifier and presentation-order sensitivity control showed no statistically robust evidence that fixed proposition encoding explained the observed condition contrasts. Conclusions: Adversarial causal influence in multi-agent LLM systems is selective rather than uniformly propagative and depends strongly on functional role and model–role configuration. Within the tested architecture, a final LLM auditor cannot be assumed to provide an independent safeguard against upstream contamination. Reliable evaluation therefore requires separate assessment of causal correctness, propagation, semantic provenance, and final-stage adversarial awareness rather than reliance on final answer accuracy alone.
- 10 October 2026
- Published by MDPI AG in Machine Learning and Knowledge Extraction
- Vol. 8 (10) , 327
Open AccessClaimArena-LLM: Auditing Adversarial Propagation of False Causal Attributions in Multi-Agent LLM Debates
Background: Multi-agent large language model (LLM) systems increasingly rely on structured interaction, debate, and final model-based evaluation, yet an unsupported causal attribution introduced by one participant may propagate through the interaction and influence subsequent decisions. This study introduces ClaimArena-LLM, a controlled framework for auditing adversarial propagation, evidentiary provenance, and final-stage evaluator vulnerability in multi-agent causal reasoning. Methods: ClaimArena-LLM uses 100 synthetic causal incident cases with predefined evidence-supported and false causal propositions. Each case was evaluated under matched adversarial and all-honest debate conditions using four functional roles: investigator, honest analyst, adversarial analyst, and auditor. The experimental program comprised 1460 validated runs, including the 200-run main experiment, claim-source lineage visibility and debate-round ablations, an 800-run balanced model–role rotation, and a 60-run proposition-identifier and presentation-order sensitivity control. Outcomes were scored deterministically from structured causal propositions, evidence references, claim lineage, reports, and auditor verdicts. Results: The adversarial analyst maintained the predefined false causal target with 97.6% mean fidelity. Adversarial exposure increased investigator shifts toward the false target by 16.0 percentage points (Holm-adjusted p = 0.035). At the auditor stage, false-target attribution increased by 11.0 percentage points (adjusted p = 0.043), verdict contamination by 20.0 percentage points (adjusted p = 0.002), and semantic provenance validity decreased by 20.0 percentage points (adjusted p = 0.002). Neither the manipulation nor the false target was explicitly detected. The ablations showed no robust adjusted differences, model–role rotation revealed substantial configuration dependence, and no benchmark-characteristic interaction remained significant after Holm correction. The proposition-identifier and presentation-order sensitivity control showed no statistically robust evidence that fixed proposition encoding explained the observed condition contrasts. Conclusions: Adversarial causal influence in multi-agent LLM systems is selective rather than uniformly propagative and depends strongly on functional role and model–role configuration. Within the tested architecture, a final LLM auditor cannot be assumed to provide an independent safeguard against upstream contamination. Reliable evaluation therefore requires separate assessment of causal correctness, propagation, semantic provenance, and final-stage adversarial awareness rather than reliance on final answer accuracy alone.
- Open Access
Exploratory Machine Learning Identification of Preoperative Laboratory Patterns Associated with Periprosthetic Joint Infection After Total Joint Arthroplasty
Background: Periprosthetic joint infection (PJI) remains one of the most severe complications following total hip and knee arthroplasty. Early identification of patients at increased risk remains challenging, particularly when routine laboratory biomarkers are within normal reference ranges. This study investigated whether routine preoperative laboratory biomarkers could reveal interpretable patterns associated with subsequent PJI using medically guided feature engineering and transparent machine-learning methods. Methods: We conducted a retrospective case–control study including 152 patients undergoing primary total hip or knee arthroplasty, comprising 75 PJI cases and 77 aseptic controls. PJI was defined according to International Consensus Meeting criteria, and aseptic controls had at least 12 months of infection-free follow-up. Thirty-eight routine preoperative laboratory biomarkers obtained one day before surgery were harmonized and analyzed. Original markers, medically derived ratios, clinically guided interaction and polynomial terms, and LASSO-based penalized selection were evaluated using 20 repetitions of nested stratified 5-fold cross-validation, with all data-dependent preprocessing and model selection confined to the corresponding training partitions. Results: Original laboratory markers showed modest discrimination (ROC-AUC 0.554; 95% CI, 0.461–0.646), while medically derived ratios performed similarly (ROC-AUC 0.556; 95% CI, 0.464–0.650). Clinically guided interaction features achieved the highest discrimination (ROC-AUC 0.643; 95% CI, 0.555–0.730), with a Brier score of 0.235 and a calibration slope of 0.349. Compared with the original-marker model, the interaction model improved ROC-AUC by 0.089 (95% CI, 0.017–0.159; p = 0.016), although this difference did not remain statistically significant after Holm adjustment (adjusted p = 0.066). Direct LASSO showed lower discrimination (ROC-AUC 0.582; 95% CI, 0.491–0.674). Conclusions: Routine preoperative laboratory biomarkers contained a modest PJI-associated signal, with clinically guided interactions yielding the highest discrimination among the evaluated laboratory representations. Overall performance remained modest, and penalized feature selection did not improve discrimination, supporting its interpretation primarily as a stability-assessment procedure. These findings should be considered exploratory and require external validation in larger, independent, preferably multicenter cohorts before clinical application.
Open AccessExploratory Machine Learning Identification of Preoperative Laboratory Patterns Associated with Periprosthetic Joint Infection After Total Joint Arthroplasty
Background: Periprosthetic joint infection (PJI) remains one of the most severe complications following total hip and knee arthroplasty. Early identification of patients at increased risk remains challenging, particularly when routine laboratory biomarkers are within normal reference ranges. This study investigated whether routine preoperative laboratory biomarkers could reveal interpretable patterns associated with subsequent PJI using medically guided feature engineering and transparent machine-learning methods. Methods: We conducted a retrospective case–control study including 152 patients undergoing primary total hip or knee arthroplasty, comprising 75 PJI cases and 77 aseptic controls. PJI was defined according to International Consensus Meeting criteria, and aseptic controls had at least 12 months of infection-free follow-up. Thirty-eight routine preoperative laboratory biomarkers obtained one day before surgery were harmonized and analyzed. Original markers, medically derived ratios, clinically guided interaction and polynomial terms, and LASSO-based penalized selection were evaluated using 20 repetitions of nested stratified 5-fold cross-validation, with all data-dependent preprocessing and model selection confined to the corresponding training partitions. Results: Original laboratory markers showed modest discrimination (ROC-AUC 0.554; 95% CI, 0.461–0.646), while medically derived ratios performed similarly (ROC-AUC 0.556; 95% CI, 0.464–0.650). Clinically guided interaction features achieved the highest discrimination (ROC-AUC 0.643; 95% CI, 0.555–0.730), with a Brier score of 0.235 and a calibration slope of 0.349. Compared with the original-marker model, the interaction model improved ROC-AUC by 0.089 (95% CI, 0.017–0.159; p = 0.016), although this difference did not remain statistically significant after Holm adjustment (adjusted p = 0.066). Direct LASSO showed lower discrimination (ROC-AUC 0.582; 95% CI, 0.491–0.674). Conclusions: Routine preoperative laboratory biomarkers contained a modest PJI-associated signal, with clinically guided interactions yielding the highest discrimination among the evaluated laboratory representations. Overall performance remained modest, and penalized feature selection did not improve discrimination, supporting its interpretation primarily as a stability-assessment procedure. These findings should be considered exploratory and require external validation in larger, independent, preferably multicenter cohorts before clinical application.
- Open Access
- 10 September 2026
- Published by MDPI AG in Informatics
- Vol. 13 (9) , 149
RubricAdapt-LLM: Measuring Criterion-Level Adaptation and Score Shifts Under Alternative Grading Rubrics
Background: Large language models are increasingly used for rubric-based grading, but it remains unclear whether they adapt selectively when criterion weights change while the response, criterion definitions, and total score remain fixed. This study examined whether alternative point allocations produce targeted criterion-level adaptation or broader grading instability. Methods: A controlled paired design was applied to 1000 student responses, comprising 500 technical and 500 argumentative answers. Eight local open-weight LLMs evaluated every response under two analytic rubrics totaling 10 points. One point was transferred from Clarity to Completeness for technical responses and from Clarity to Dialecticality for argumentative responses. Of 16,000 expected evaluations, 15,999 were structurally valid, yielding 7999 complete cross-rubric pairs. Model outputs were analyzed for total-score shifts, affected-criterion adaptation, stability of unaffected criteria, model–human alignment, and correspondence with differences between two human evaluation conditions. Results: All models assigned lower mean scores under Rubric B, with mean shifts ranging from −1.239 to −0.182 points. Adaptation mechanisms differed substantially across models and response types. gemma3:4b frequently preserved technical total scores through compensating criterion changes, whereas llama3.1:8b showed extensive spillover into unaffected criteria. The Qwen models generally produced smaller total-score reductions and greater stability in unchanged dimensions. The mean score was 0.637 points higher under the human Rubric B condition than under the human Rubric A condition, although the two conditions were applied by different evaluator pairs; every model shifted negatively, and the lowest overall shift error was obtained by qwen3:4b at 1.498 points. Conclusions: Rubric sensitivity did not consistently imply localized or criterion-consistent adaptation. Reliable evaluation of LLM graders therefore requires separate analysis of total scores, affected criteria, unaffected criteria, human alignment, and cross-condition shift correspondence.
- 10 September 2026
- Published by MDPI AG in Informatics
- Vol. 13 (9) , 149
Open AccessRubricAdapt-LLM: Measuring Criterion-Level Adaptation and Score Shifts Under Alternative Grading Rubrics
Background: Large language models are increasingly used for rubric-based grading, but it remains unclear whether they adapt selectively when criterion weights change while the response, criterion definitions, and total score remain fixed. This study examined whether alternative point allocations produce targeted criterion-level adaptation or broader grading instability. Methods: A controlled paired design was applied to 1000 student responses, comprising 500 technical and 500 argumentative answers. Eight local open-weight LLMs evaluated every response under two analytic rubrics totaling 10 points. One point was transferred from Clarity to Completeness for technical responses and from Clarity to Dialecticality for argumentative responses. Of 16,000 expected evaluations, 15,999 were structurally valid, yielding 7999 complete cross-rubric pairs. Model outputs were analyzed for total-score shifts, affected-criterion adaptation, stability of unaffected criteria, model–human alignment, and correspondence with differences between two human evaluation conditions. Results: All models assigned lower mean scores under Rubric B, with mean shifts ranging from −1.239 to −0.182 points. Adaptation mechanisms differed substantially across models and response types. gemma3:4b frequently preserved technical total scores through compensating criterion changes, whereas llama3.1:8b showed extensive spillover into unaffected criteria. The Qwen models generally produced smaller total-score reductions and greater stability in unchanged dimensions. The mean score was 0.637 points higher under the human Rubric B condition than under the human Rubric A condition, although the two conditions were applied by different evaluator pairs; every model shifted negatively, and the lowest overall shift error was obtained by qwen3:4b at 1.498 points. Conclusions: Rubric sensitivity did not consistently imply localized or criterion-consistent adaptation. Reliable evaluation of LLM graders therefore requires separate analysis of total scores, affected criteria, unaffected criteria, human alignment, and cross-condition shift correspondence.
- Open Access
CreditTrace-LLM: Auditing Rubric-Point Responsiveness and Credit Locality in LLM-Based Automated Grading
Large language models are increasingly used for automated grading, but final-score agreement does not reveal whether credit is assigned to the rubric element affected by a response change. This study introduces CreditTrace-LLM, a controlled framework for auditing directional responsiveness, credit locality, paraphrase stability, and correspondence with expert evaluations in rubric-guided grading. The evaluation used 500 response families across ten questions, equally divided between technical and argumentative tasks. Each family included a baseline response (C0), a positive intervention adding rubric-relevant evidence (C+), a negative intervention removing or weakening such evidence (C−), and a placebo paraphrase (CP) constructed through lexical or syntactic reformulation with the intention of preserving rubric-relevant meaning. Eight local open-weight LLM graders produced 15,999 structurally valid Gold-point-level outputs from 16,000 expected evaluations. Targeted Gold-point scores changed in the expected direction in 75.7% of C+ comparisons and 50.2% of C− comparisons. Localized directional success was lower under C+ and C−, at 31.5% and 21.3%, respectively, indicating frequent non-target score changes. Under CP, total-score stability was 75.4%, while complete-profile stability was 73.6%. Direction concordance with the mean expert score change was 69.8% for C+, 47.4% for C−, and 18.2% for CP. The CP condition also showed substantial variability in expert scoring, particularly for argumentative responses, indicating that intended rubric-relevant meaning preservation did not guarantee score invariance. Final-score behavior alone is insufficient for validating rubric-guided LLM grading. CreditTrace-LLM therefore evaluates target responsiveness, credit locality, paraphrase stability, and traceable Gold-point-level outputs as complementary diagnostic dimensions rather than as predefined criteria for acceptable grading performance.
Open AccessCreditTrace-LLM: Auditing Rubric-Point Responsiveness and Credit Locality in LLM-Based Automated Grading
Large language models are increasingly used for automated grading, but final-score agreement does not reveal whether credit is assigned to the rubric element affected by a response change. This study introduces CreditTrace-LLM, a controlled framework for auditing directional responsiveness, credit locality, paraphrase stability, and correspondence with expert evaluations in rubric-guided grading. The evaluation used 500 response families across ten questions, equally divided between technical and argumentative tasks. Each family included a baseline response (C0), a positive intervention adding rubric-relevant evidence (C+), a negative intervention removing or weakening such evidence (C−), and a placebo paraphrase (CP) constructed through lexical or syntactic reformulation with the intention of preserving rubric-relevant meaning. Eight local open-weight LLM graders produced 15,999 structurally valid Gold-point-level outputs from 16,000 expected evaluations. Targeted Gold-point scores changed in the expected direction in 75.7% of C+ comparisons and 50.2% of C− comparisons. Localized directional success was lower under C+ and C−, at 31.5% and 21.3%, respectively, indicating frequent non-target score changes. Under CP, total-score stability was 75.4%, while complete-profile stability was 73.6%. Direction concordance with the mean expert score change was 69.8% for C+, 47.4% for C−, and 18.2% for CP. The CP condition also showed substantial variability in expert scoring, particularly for argumentative responses, indicating that intended rubric-relevant meaning preservation did not guarantee score invariance. Final-score behavior alone is insufficient for validating rubric-guided LLM grading. CreditTrace-LLM therefore evaluates target responsiveness, credit locality, paraphrase stability, and traceable Gold-point-level outputs as complementary diagnostic dimensions rather than as predefined criteria for acceptable grading performance.
- Open Access
- 25 August 2026
- Published by MDPI AG in Applied System Innovation
- Vol. 9 (9) , 174
ConsensusGrade: A Human-Variability-Aware Framework for Evaluating LLM-Based Automated Grading
Background: Evaluation of LLM-based automated grading often relies on comparison with a single human score, which can obscure meaningful variability among raters of open-ended answers. This study introduces ConsensusGrade, a consensus-aware framework that treats the human reference as a scoring envelope rather than as a single point. Methods: We analyzed 1000 open-ended student answers from 100 students across 10 questions, each graded by four evaluators. Six previously generated and aligned automated grading configurations from GradeAgentOps were compared with the four-rater human reference. The score sets were generated using Llama 3.3 70B Instruct as the primary grader, with Qwen 2.5 14B Instruct for semantic repair. Results: Human evaluators showed meaningful agreement, with ICC(A,1) = 0.712, but exact four-rater agreement occurred in only 2.2% of records. Broad score dispersion occurred in 59.0%. All automated configurations showed negative bias relative to the human median. FULL achieved 68.5% inside-envelope positioning and a chance-adjusted score of 0.454; under the central-trimmed envelope, this rate decreased to 34.3%, while configuration ordering was preserved. Conclusions: ConsensusGrade provides a diagnostic framework for interpreting automated scores relative to observed human variability; inside-envelope rates should not be interpreted as stand-alone measures of grading accuracy.
- 25 August 2026
- Published by MDPI AG in Applied System Innovation
- Vol. 9 (9) , 174
Open AccessConsensusGrade: A Human-Variability-Aware Framework for Evaluating LLM-Based Automated Grading
Background: Evaluation of LLM-based automated grading often relies on comparison with a single human score, which can obscure meaningful variability among raters of open-ended answers. This study introduces ConsensusGrade, a consensus-aware framework that treats the human reference as a scoring envelope rather than as a single point. Methods: We analyzed 1000 open-ended student answers from 100 students across 10 questions, each graded by four evaluators. Six previously generated and aligned automated grading configurations from GradeAgentOps were compared with the four-rater human reference. The score sets were generated using Llama 3.3 70B Instruct as the primary grader, with Qwen 2.5 14B Instruct for semantic repair. Results: Human evaluators showed meaningful agreement, with ICC(A,1) = 0.712, but exact four-rater agreement occurred in only 2.2% of records. Broad score dispersion occurred in 59.0%. All automated configurations showed negative bias relative to the human median. FULL achieved 68.5% inside-envelope positioning and a chance-adjusted score of 0.454; under the central-trimmed envelope, this rate decreased to 34.3%, while configuration ordering was preserved. Conclusions: ConsensusGrade provides a diagnostic framework for interpreting automated scores relative to observed human variability; inside-envelope rates should not be interpreted as stand-alone measures of grading accuracy.
- Open Access
- 21 August 2026
- Published by MDPI AG in Machine Learning and Knowledge Extraction
- Vol. 8 (8) , 252
GradeDrift-LLM: Measuring Student-History-Induced Score Drift in LLM-Based Automated Grading
Background: Large language models (LLMs) are increasingly explored for automated educational assessment, while future educational platforms may combine grading, feedback, learner analytics, and personalization. The objective of this study was to determine whether student-history metadata can influence the numerical score assigned to the same answer. Methods: This study introduces GradeDrift-LLM, a controlled framework for measuring student-history-induced score drift in LLM-based automated grading. We evaluated 1000 Computer Science answers from 100 students across six student-history conditions and eight open-weight LLMs. For each grading instance, the submitted answer, question, reference answer, rubric-related information, scoring scale, and grading instruction were kept constant; only the student-history condition varied. Results: Across 39,997 valid paired comparisons, 83.92% showed no drift, 9.40% showed upward drift, and 6.68% showed downward drift. Mean absolute drift was 0.2137 points, and the 95th percentile absolute drift was 1 point. Positive-history frames tended to increase scores, whereas negative-history frames tended to decrease them. Drift was model-dependent, not uniformly explained by approximate scale, and present in both technical and argumentative answers; rare extreme deviations reached 10 points. Conclusions: Student-history metadata can influence LLM-generated grading scores despite explicit instructions to ignore it. Future LLM-based grading systems should separate answer-based scoring from learner-context-based personalization and validate score invariance under controlled learner-context variations.
- 21 August 2026
- Published by MDPI AG in Machine Learning and Knowledge Extraction
- Vol. 8 (8) , 252
Open AccessGradeDrift-LLM: Measuring Student-History-Induced Score Drift in LLM-Based Automated Grading
Background: Large language models (LLMs) are increasingly explored for automated educational assessment, while future educational platforms may combine grading, feedback, learner analytics, and personalization. The objective of this study was to determine whether student-history metadata can influence the numerical score assigned to the same answer. Methods: This study introduces GradeDrift-LLM, a controlled framework for measuring student-history-induced score drift in LLM-based automated grading. We evaluated 1000 Computer Science answers from 100 students across six student-history conditions and eight open-weight LLMs. For each grading instance, the submitted answer, question, reference answer, rubric-related information, scoring scale, and grading instruction were kept constant; only the student-history condition varied. Results: Across 39,997 valid paired comparisons, 83.92% showed no drift, 9.40% showed upward drift, and 6.68% showed downward drift. Mean absolute drift was 0.2137 points, and the 95th percentile absolute drift was 1 point. Positive-history frames tended to increase scores, whereas negative-history frames tended to decrease them. Drift was model-dependent, not uniformly explained by approximate scale, and present in both technical and argumentative answers; rare extreme deviations reached 10 points. Conclusions: Student-history metadata can influence LLM-generated grading scores despite explicit instructions to ignore it. Future LLM-based grading systems should separate answer-based scoring from learner-context-based personalization and validate score invariance under controlled learner-context variations.
- Open Access
- 14 August 2026
- Published by MDPI AG in Journal of Clinical Medicine
- Vol. 15 (16) , 6297
From Medical Records to AI-Ready Datasets: A Practical Guide for Clinical Researchers
Background: Medical artificial intelligence (AI), machine learning (ML), and deep learning (DL) studies frequently begin with datasets collected for routine care rather than for computational modeling. Such datasets may contain inconsistent variables, heterogeneous measurement time points, unexplained NaN values, poorly defined outcomes, missing metadata, and insufficient documentation, which can compromise model development before any algorithm is selected. Methods: This Technical Note proposes a physician-facing Clinical AI-Readiness Guide for preparing medical datasets before AI-based analysis. The guide was developed as a practical framework organized around pre-modeling decisions, including the clinical task, cohort, minimum common dataset, outcome definition, predictor variables, measurement timing, missing-data logic, standardization, non-tabular data linkage, data dictionary, and validation readiness. Results: The proposed guide translates AI-readiness principles into concrete data-collection rules for clinical, laboratory, imaging, physiological-signal, textual, follow-up, and multimodal data. It emphasizes clinically consistent data acquisition, reliable target labeling, explicit missing-data logic, patient-level linkage, structured metadata, and validation feasibility. A structured checklist and scoring approach are also proposed as practical pre-modeling assessment tools to classify datasets as not ready, exploratory only, ML-ready with limitations, or AI-ready for model development. Conclusions: Medical AI-readiness should be established before model development begins. By helping physicians collect, structure, and document data more consistently, the proposed guide may improve collaboration between clinical and technical teams and reduce preventable dataset-related failures in medical AI research.
- 14 August 2026
- Published by MDPI AG in Journal of Clinical Medicine
- Vol. 15 (16) , 6297
Open AccessFrom Medical Records to AI-Ready Datasets: A Practical Guide for Clinical Researchers
Background: Medical artificial intelligence (AI), machine learning (ML), and deep learning (DL) studies frequently begin with datasets collected for routine care rather than for computational modeling. Such datasets may contain inconsistent variables, heterogeneous measurement time points, unexplained NaN values, poorly defined outcomes, missing metadata, and insufficient documentation, which can compromise model development before any algorithm is selected. Methods: This Technical Note proposes a physician-facing Clinical AI-Readiness Guide for preparing medical datasets before AI-based analysis. The guide was developed as a practical framework organized around pre-modeling decisions, including the clinical task, cohort, minimum common dataset, outcome definition, predictor variables, measurement timing, missing-data logic, standardization, non-tabular data linkage, data dictionary, and validation readiness. Results: The proposed guide translates AI-readiness principles into concrete data-collection rules for clinical, laboratory, imaging, physiological-signal, textual, follow-up, and multimodal data. It emphasizes clinically consistent data acquisition, reliable target labeling, explicit missing-data logic, patient-level linkage, structured metadata, and validation feasibility. A structured checklist and scoring approach are also proposed as practical pre-modeling assessment tools to classify datasets as not ready, exploratory only, ML-ready with limitations, or AI-ready for model development. Conclusions: Medical AI-readiness should be established before model development begins. By helping physicians collect, structure, and document data more consistently, the proposed guide may improve collaboration between clinical and technical teams and reduce preventable dataset-related failures in medical AI research.
- Open Access
SafetyJudge-LLM: Auditing Local Open-Weight LLMs as Semantic Safety Judges for Boundary-Failure Detection
Background: LLM-as-a-judge workflows are increasingly used to evaluate open-ended model outputs, but the judge model can itself become a source of error in safety assessment. SafetyJudge-LLM audits local open-weight LLMs as semantic safety judges. Methods: This study reused a fixed set of previously reviewed safety-boundary responses and their hidden reference labels. Two independent human evaluations (R1 and R2) quantified reference-layer ambiguity. Seven local open-weight judge models were evaluated under a common Ollama inference protocol. A paired C6 sensitivity analysis reran llama3.2:3b and qwen3:8b through Hugging Face Transformers. Results: The final judge-output matrix contained 10,612 retained outputs. R1–R2 agreement was 95.45% (Cohen’s κ = 0.612) overall but 47.80% (κ = 0.341) in secondary cases. Several judge models detected more than 90% of confirmed safety-boundary failures, but high detection was not always accompanied by low false-unsafe behavior on control cases. Output-format reliability also varied across models: overall label parseability was 98.11%, while strict JSON schema compliance was 92.55%. The llama3.2:3b schema-failure rate persisted across engines (52.06% under Ollama; 59.60% under Transformers), whereas qwen3:8b maintained complete compliance. Conclusions: SafetyJudge-LLM shows that local open-weight LLMs can support semantic safety judging, but their reliability must be evaluated across multiple dimensions.
Open AccessSafetyJudge-LLM: Auditing Local Open-Weight LLMs as Semantic Safety Judges for Boundary-Failure Detection
Background: LLM-as-a-judge workflows are increasingly used to evaluate open-ended model outputs, but the judge model can itself become a source of error in safety assessment. SafetyJudge-LLM audits local open-weight LLMs as semantic safety judges. Methods: This study reused a fixed set of previously reviewed safety-boundary responses and their hidden reference labels. Two independent human evaluations (R1 and R2) quantified reference-layer ambiguity. Seven local open-weight judge models were evaluated under a common Ollama inference protocol. A paired C6 sensitivity analysis reran llama3.2:3b and qwen3:8b through Hugging Face Transformers. Results: The final judge-output matrix contained 10,612 retained outputs. R1–R2 agreement was 95.45% (Cohen’s κ = 0.612) overall but 47.80% (κ = 0.341) in secondary cases. Several judge models detected more than 90% of confirmed safety-boundary failures, but high detection was not always accompanied by low false-unsafe behavior on control cases. Output-format reliability also varied across models: overall label parseability was 98.11%, while strict JSON schema compliance was 92.55%. The llama3.2:3b schema-failure rate persisted across engines (52.06% under Ollama; 59.60% under Transformers), whereas qwen3:8b maintained complete compliance. Conclusions: SafetyJudge-LLM shows that local open-weight LLMs can support semantic safety judging, but their reliability must be evaluated across multiple dimensions.
- Open Access
SafeBoundary-LLM: Measuring Safety Boundary Stability in Local Open-Weight LLMs Through Single-Turn Baselines and Multi-Turn Escalation
Local open-weight large language models (LLMs) are increasingly used in privacy-sensitive settings, yet isolated prompts may not reveal whether safety boundaries remain stable during conversation. SafeBoundary-LLM evaluated seven local models across 14 sensitive domains, 84 boundary sets, 672 single-turn prompts, and 84 five-turn escalation conversations; the same models were evaluated separately on XSTest and JBB-Behaviors. Evaluators R1 and R2 independently classified all 12,194 responses, with R2 labels used for primary outcomes and unreconciled labels used for reliability analysis. Exact agreement exceeded 93% in each dataset. In SafeBoundary-LLM, 456 out of 7644 responses (5.97%) were confirmed-or-mixed failures. The multi-turn failure rate was 14.69% versus 0.51% for single-turn prompts, yielding a rate ratio of 28.80 (95% CI [21.20, 43.80]; Holm-adjusted p = 0.0006); boundary collapse occurred only at Turns 4–5, and role-play bypass accounted for 299 out of 456 failures. On answer-expected items, over-refusal was 4.34% in XSTest and 17.29% in JBB-Behaviors, whereas unsafe compliance on refusal-expected items was 0.36% and 0.86%, respectively. These findings support an evaluation strategy that includes public single-turn benchmarks, controlled multi-turn escalation, independent human review, and traceable audit records for locally deployed LLMs.
Open AccessSafeBoundary-LLM: Measuring Safety Boundary Stability in Local Open-Weight LLMs Through Single-Turn Baselines and Multi-Turn Escalation
Local open-weight large language models (LLMs) are increasingly used in privacy-sensitive settings, yet isolated prompts may not reveal whether safety boundaries remain stable during conversation. SafeBoundary-LLM evaluated seven local models across 14 sensitive domains, 84 boundary sets, 672 single-turn prompts, and 84 five-turn escalation conversations; the same models were evaluated separately on XSTest and JBB-Behaviors. Evaluators R1 and R2 independently classified all 12,194 responses, with R2 labels used for primary outcomes and unreconciled labels used for reliability analysis. Exact agreement exceeded 93% in each dataset. In SafeBoundary-LLM, 456 out of 7644 responses (5.97%) were confirmed-or-mixed failures. The multi-turn failure rate was 14.69% versus 0.51% for single-turn prompts, yielding a rate ratio of 28.80 (95% CI [21.20, 43.80]; Holm-adjusted p = 0.0006); boundary collapse occurred only at Turns 4–5, and role-play bypass accounted for 299 out of 456 failures. On answer-expected items, over-refusal was 4.34% in XSTest and 17.29% in JBB-Behaviors, whereas unsafe compliance on refusal-expected items was 0.36% and 0.86%, respectively. These findings support an evaluation strategy that includes public single-turn benchmarks, controlled multi-turn escalation, independent human review, and traceable audit records for locally deployed LLMs.
- Open Access
- 18 June 2026
- Published by MDPI AG in Journal of Imaging
- Vol. 12 (6) , 270
Cutaneous Thermography in Arthropathies: Quantitative Imaging, Machine Learning, and Clinical Translation
Arthropathies are a major global health challenge because of their high prevalence, chronic progression, and significant impact on quality of life and health systems. Therefore, prompt and accurate diagnosis is critical for slowing disease progression and improving outcomes. Traditional imaging modalities, such as ultrasound and magnetic resonance imaging, suffer from significant limitations, including operator dependence, limited accessibility, high cost, and limited reproducibility. Infrared thermography has become a promising non-invasive imaging technique for identifying thermal variations linked to inflammatory and metabolic processes. Advances in quantitative thermography, automated segmentation, and artificial intelligence have greatly enhanced its clinical applicability. This review summarizes recent advances in thermography-based biomarkers, including region-of-interest-derived metrics, asymmetry indices, hotspot burden, spatial and texture descriptors, and composite thermographic scores. It discusses the role of machine learning and deep learning in prediction, phenotyping, and multimodal integration with clinical, laboratory, and imaging data. Heterogeneity of protocols, variability in measurements, domain shift, validation design, overfitting, and reporting quality are also addressed. Overall, thermography combined with AI is highly promising as an adjunct to early diagnosis, assessment of disease activity, and follow-up in arthropathies. However, clinical application at a large scale requires strict standardization, external validation, transparent reporting, and well-elucidated, reproducible analytical processes.
- 18 June 2026
- Published by MDPI AG in Journal of Imaging
- Vol. 12 (6) , 270
Open AccessCutaneous Thermography in Arthropathies: Quantitative Imaging, Machine Learning, and Clinical Translation
Arthropathies are a major global health challenge because of their high prevalence, chronic progression, and significant impact on quality of life and health systems. Therefore, prompt and accurate diagnosis is critical for slowing disease progression and improving outcomes. Traditional imaging modalities, such as ultrasound and magnetic resonance imaging, suffer from significant limitations, including operator dependence, limited accessibility, high cost, and limited reproducibility. Infrared thermography has become a promising non-invasive imaging technique for identifying thermal variations linked to inflammatory and metabolic processes. Advances in quantitative thermography, automated segmentation, and artificial intelligence have greatly enhanced its clinical applicability. This review summarizes recent advances in thermography-based biomarkers, including region-of-interest-derived metrics, asymmetry indices, hotspot burden, spatial and texture descriptors, and composite thermographic scores. It discusses the role of machine learning and deep learning in prediction, phenotyping, and multimodal integration with clinical, laboratory, and imaging data. Heterogeneity of protocols, variability in measurements, domain shift, validation design, overfitting, and reporting quality are also addressed. Overall, thermography combined with AI is highly promising as an adjunct to early diagnosis, assessment of disease activity, and follow-up in arthropathies. However, clinical application at a large scale requires strict standardization, external validation, transparent reporting, and well-elucidated, reproducible analytical processes.