Codruta Istin

Introduction: Therapeutic plasma exchange (TPE) is increasingly used as an adjunctive intervention in severe, hyperinflammatory critical illness, including COVID-19, yet clinical guidance remains syndromic and evidence is heterogeneous. We present an integrated, interpretable machine-learning framework designed to support protocolizable TPE decision-making by identifying biochemical phenotypes associated with short-horizon laboratory response to TPE. Our dataset consists of real-world intensive care unit cases and captures the treatment heterogeneity and operational constraints that a workable institutional protocol must accommodate, being well-suited as a “protocol seed” for iterative validation. Methods: We jointly analyze a COVID-19 cohort and a non-COVID comparator cohort receiving TPE. Three decision trees were constructed to represent: (1) global biochemical improvement, (2) strict improvement dependent on key inflammatory/coagulation markers, and (3) early interleukin-6 (IL-6) response. The models revealed distinct favorable phenotypes—particularly patients with IL-6 > 86 pg/mL, lactate dehydrogenase (LDH) >346 U/L, lymphopenia, and fibrinogen ≤ 8.3 g/L. Apriori analysis further identified high-confidence patterns linking high values of IL-6 and LDH with TPE responsiveness. We constructed a unified four-tier clinical algorithm for candidate stratification. Results: The resulting logic aligns with how TPE has been applied in practice in published severe COVID-19 series and trials, where candidate selection typically targets cytokine release syndrome-like phenotypes, organ dysfunction, and hyperinflammatory biomarker profiles. We further contextualize these findings against American Society for Apheresis guidance and COVID-era operational considerations, demonstrating convergence while adding quantitative biomarker thresholds. Conclusion: These findings support two complementary perspectives of benefit: (i) an algorithmic framework that provides transparent, protocol-ready guidelines, and (ii) a phenotype-based clinical approach consistent with observed post-TPE marker changes in severe COVID-19 cases. Larger multicenter validation is warranted; however, the present work provides a practical foundation for protocol construction and auditable decision support in settings already performing TPE.

In the context of unsupervised learning, effective clustering plays a vital role in revealing patterns and insights from unlabeled data. However, the success of clustering algorithms often depends on the relevance and contribution of features, which can differ between various datasets. This paper explores feature weighting for clustering and presents new weighting strategies, including methods based on SHAP (SHapley Additive exPlanations), a technique commonly used for providing explainability in various supervised machine learning tasks. By taking advantage of SHAP values in a way other than just to gain explainability, we use them to weight features and ultimately improve the clustering process itself in unsupervised scenarios. Our empirical evaluations across five benchmark datasets and clustering methods demonstrate that feature weighting based on SHAP can enhance unsupervised clustering quality, achieving up to a 22.69% improvement over other weighting methods (from 0.586 to 0.719 in terms of the Adjusted Rand Index). Additionally, these situations where the weighted data boosts the results are highlighted and thoroughly explored, offering insight for practical applications.

Recent advancements in Vision Transformers (ViT) have demonstrated exceptional results in various visual recognition tasks, owing to their ability to capture long-range dependencies in images through self-attention mechanisms. However, the complex nature of ViT models requires robust explainability methods to unveil their decision-making processes. Explainable Artificial Intelligence (XAI) plays a crucial role in improving model transparency and trustworthiness by providing insights into model predictions. Current approaches to ViT explainability, based on visualization techniques such as Layer-wise Relevance Propagation (LRP) and gradient-based methods, have shown promising but sometimes limited results. In this study, we explore a hybrid approach that mixes multiple explainability techniques to overcome these limitations and enhance the interpretability of ViT models. Our experiments reveal that this hybrid approach significantly improves the interpretability of ViT models compared to individual methods. We also introduce modifications to existing techniques, such as using geometric mean for mixing, which demonstrates notable results in object segmentation tasks. To quantify the explainability gain, we introduced a novel post-hoc explainability measure by applying the Pigeonhole principle. These findings underscore the importance of refining and optimizing explainability methods for ViT models, paving the way to reliable XAI-based segmentations.

In this paper, a method is introduced to control the dark knowledge values also known as soft targets, with the purpose of improving the training by knowledge distillation for multi-class classification tasks. Knowledge distillation effectively transfers knowledge from a larger model to a smaller model to achieve efficient, fast, and generalizable performance while retaining much of the original accuracy. The majority of deep neural models used for classification tasks append a SoftMax layer to generate output probabilities and it is usual to take the highest score and consider it the inference of the model, while the rest of the probability values are generally ignored. The focus is on those probabilities as carriers of dark knowledge and our aim is to quantify the relevance of dark knowledge, not heuristically as provided in the literature so far, but with an inductive proof on the SoftMax operational limits. These limits are further pushed by using an incremental decision tree with information gain split. The user can set a desired precision and an accuracy level to obtain a maximal temperature setting for a continual classification process. Moreover, by fitting both the hard targets and the soft targets, one obtains an optimal knowledge distillation effect that mitigates better catastrophic forgetting. The strengths of our method come from the possibility of controlling the amount of distillation transferred non-heuristically and the agnostic application of this model-independent study.

The purpose of this paper is to present a detailed investigation of the advantages of employing GraphLIME (Local Interpretable Model Explanations for Graph Neural Networks) for the trustworthy prediction of diabetes mellitus. Our pursuit involves identifying the strengths of GraphLIME combined with the attention-mechanism over the standard coupling of deep learning neural networks with the original LIME method. The system build this way, provided us a proficient method for extracting the most relevant features and applying the attention mechanism exclusively to those features. We have closely monitored the performance metrics of the two approaches and conducted a comparative analysis. Leveraging attention mechanisms, we have achieved an accuracy of 92.6% for the addressed problem. The model’s performance is meticulously demonstrated throughout the study, and the results are furthermore evaluated using the Receiver Operating Characteristic (ROC) curve. By implementing this technique on a dataset of 768 patients diagnosed with or without diabetes mellitus, we have successfully boosted the model’s performance by over 18%.

(1) Background: Lung cancers are the most common cancers worldwide, and prostate cancers are among the second in terms of the frequency of cancers diagnosed in men. Automatic ranking of the risk groups of such diseases is highly in demand, but the clinical practice has shown us that, for a sensitive screening of the clinical parameters using an artificial intelligence system, a customarily defined deep neural network classifier is not sufficient given the usually small size of medical datasets. (2) Methods: In this paper, we propose a new management method of cancer risk groups based on a supervised neural network model that is further enhanced by using a features attention mechanism in order to boost its level of accuracy. For the analysis of each clinical parameter, we used local interpretable model-agnostic explanations, which is a post hoc model-agnostic technique that outlines feature importance. After that, we applied the feature-attention mechanism in order to obtain a higher weight after training. We tested the method on two datasets, one for binary-class in cases of thoracic cancer and one for multi-class classification in cases of urological cancer, to demonstrate the wide availability and versatility of the method. (3) Results: The accuracy levels of the models trained in this way reached values of more than 80% for both clinical tasks. (4) Conclusions: Our experiments demonstrate that, by using explainability results as feedback signals in conjunction with the attention mechanism, we were able to increase the accuracy of the base model by more than 20% on small medical datasets, reaching a critical threshold for providing recommendations based on the collected clinical parameters.

We introduce in this paper a neuro-symbolic predictive model based on Logic Tensor Networks, capable of discriminating and at the same time of explaining the bad connections, called alerts or attacks, and the normal connections. The proposed classifier incorporates both the ability of deep neural networks to improve on their own through learning from experience and the interpretability of the results provided by the symbolic artificial intelligence approach. Compared to other existing solutions, we advance in the discovery of potential security breaches from a cognitive perspective. By introducing the reasoning in the model, our aim is to further reduce the human staff needed to deal with the cyber-threat hunting problem. To justify the need for shifting towards hybrid systems for this task, the design, the implementation, and the comparison of the dense neural network and the neuro-symbolic model is performed in detail. While in terms of standard accuracy, both models demonstrated similar precision, we further introduced for our model the concept of interactive accuracy as a way of querying the model results at any time coupled with deductive reasoning over data. By applying our model on the CIC-IDS2017 dataset, we reached an accuracy of 0.95, with levels of satisfiability around 0.85. Other advantages such as overfitting mitigation and scalability issues are also presented.