BioMedInformatics, Volume 5, Issue 4 (December 2025) – 6 articles

Background: The growing demand for automated microorganism classification in the context of Laboratory 4.0 highlights the potential of convolutional neural networks (CNNs) for accurate and efficient image analysis. However, their effectiveness remains limited by the scarcity of large, labeled datasets. This study addresses a key gap in the literature by investigating how commonly used image preprocessing techniques, such as lossy compression, non-uniform scaling (typically applied to fit input images to CNN input layers), and data augmentation, affect the performance of CNNs in automated microorganism classification. Methods: Using two well-established CNN architectures, AlexNet and DenseNet-121, both frequently applied in biomedical image analysis, we conducted a series of computational experiments on a standardized dataset of high-resolution bacterial images. Results: Our results demonstrate under which conditions these preprocessing strategies degrade or improve CNN performance. Using the findings from this research to optimize hyperparameters and train the CNNs, we achieved classification accuracies of 98.61% with AlexNet and 99.82% with DenseNet-121, surpassing the performance reported in current state-of-the-art studies. Conclusions: This study advances laboratory digitalization by reducing data preparation effort, training time, and computational costs, while improving the accuracy of microorganism classification with deep learning. Its contributions also benefit broader biomedical fields such as automated diagnostics, digital pathology, clinical decision support, and point-of-care imaging. Full article

Parkinson’s disease (PD) is the second most common neurodegenerative disorder, characterized primarily by motor impairments due to the loss of dopaminergic neurons. Despite extensive research, the precise causes of PD remain unknown, and reliable non-invasive biomarkers are still lacking. This study aimed to explore gene expression profiles in peripheral blood to identify potential biomarkers for PD using machine learning approaches. We analyzed microarray-based gene expression data from 105 individuals (50 PD patients, 33 with other neurodegenerative diseases, and 22 healthy controls) obtained from the GEO database (GSE6613). Preprocessing was performed using the “affy” package in R with Expresso normalization. Feature selection and classification were conducted using a decision tree approach (C4.5/J48 algorithm in WEKA), and model performance was evaluated with 10-fold cross-validation. Additional classifiers such as Support Vector Machine (SVM), the Naive Bayes classifier and Multilayer Perceptron Neural Network (MLP) were used for comparison. ROC curve analysis and Gene Ontology (GO) enrichment analysis were applied to the selected genes. A nine-gene decision tree model (TMEM104, TRIM33, GJB3, SPON2, SNAP25, TRAK2, SHPK, PIEZO1, RPL37) achieved 86.71% accuracy, 88% sensitivity, and 87% specificity. The model significantly outperformed other classifiers (SVM, Naive Bayes, MLP) in terms of overall predictive accuracy. ROC analysis showed moderate discrimination for some genes (e.g., TRAK2, TRIM33, PIEZO1), and GO enrichment revealed associations with synaptic processes, inflammation, mitochondrial transport, and stress response pathways. Our decision tree model based on blood gene expression profiles effectively discriminates between PD, other neurodegenerative conditions, and healthy controls, offering a non-invasive method for potential early diagnosis. Notably, TMEM104, TRIM33, and SNAP25 emerged as promising candidate biomarkers, warranting further investigation in larger and synthetic datasets to validate their clinical relevance. Full article

Background: Reconstructing gene regulatory networks (GRNs) from gene expression data remains a major challenge in systems biology due to the inherent complexity of biological systems and the limitations of existing reconstruction methods, which often yield high false-positive rates. This study aims to develop a robust and adaptive approach to enhance the accuracy of inferred GRNs by integrating multiple modelling paradigms. Methods: We introduce EvoFuzzy, a novel algorithm that integrates evolutionary computation and fuzzy logic to aggregate GRNs reconstructed using Boolean, regression, and fuzzy modelling techniques. The algorithm initializes an equal number of individuals from each modelling method to form a diverse population, which evolves through fuzzy trigonometric differential evolution. Gene expression values are predicted using a fuzzy logic-based predictor with confidence levels, and a fitness function is applied to identify the optimal consensus network. Results: The proposed method was evaluated using simulated benchmark datasets and a real-world SOS gene repair dataset. Experimental results demonstrated that EvoFuzzy consistently outperformed existing state-of-the-art GRN reconstruction methods in terms of accuracy and robustness. Conclusions: The fuzzy trigonometric differential evolution approach plays a pivotal role in refining and aggregating multiple network outputs into a single, optimal consensus network, making EvoFuzzy a powerful and reliable framework for reconstructing biologically meaningful gene regulatory networks. Full article

Extensive research is being carried out on the application of $\alpha$ particles for cancer treatment. A key challenge in $\alpha$ therapy is how to deliver the $\alpha$ emitters to the tumour. In AlphaGlue, a novel treatment delivery concept, the $\alpha$ emitters are suspended in a thin layer of glue that is put on top of the tumour. In principle, this should be an easy and safe way to apply $\alpha$ therapy. In this study, the effectiveness of AlphaGlue is evaluated using GEANT4 and GEANT4-DNA simulations to calculate the DNA damage as a function of depth. Two radionuclides are considered in this work, ²¹¹At and ²²⁴Ra. The results indicate that, as a concept, the method offers a promising hypothesis for treating superficial tumours, such as skin cancer, when ²²⁴Ra is applied directly on the tissue and stabilized with a glue layer. This results in 2$\times {10}^{-5}$ complex double strand breaks and 5$\times {10}^{-5}$ double strand breaks at 5 mm depth per applied ²²⁴Ra atom. When applying a ²²⁴Ra atom concentration of $(4.35\pm 0.2)\times {10}^{11}/$cm² corresponding to an activity of $(21.8\pm 1)\mathsf{\mu }$Ci/cm² on the skin surface, the RBE weighted dose exceeds 20 Gy at 5 mm depth. Hence, there is significant cell death at 5 mm into the tissue; a depth matching clinical requirements for skin cancer treatment. Given the rapidly falling weighted dose versus depth curve, the treatment depth can be tuned with good precision. The results of this study show that AlphaGlue is a promosing treatment and open the pathway towards the next stage of the research, which includes in-vitro studies. Full article

Background: Urinary tract infections (UTIs) remain among the most common bacterial infections, yet reliable risk stratification remains challenging. Serum vitamin D has been linked to immune regulation, but its predictive role in UTI subtypes is unclear. Methods: We analyzed 332 de-identified clinical records using six machine learning algorithms: Extra Trees, Gradient Boosting, XGBoost, Logistic Regression, Random Forest, and LightGBM. Two preprocessing strategies were applied: (i) removing rows with missing fasting blood sugar (FBs) and HbA1c, and (ii) dropping columns with Null FBs and HbA1c values. Model performance was evaluated using 10-fold cross-validation. Results: Serum vitamin D showed weak correlations with UTI subtypes but modest importance in tree-based models. The highest predictive accuracy was obtained with Extra Trees (0.9510) under the row-removal strategy and Random Forest (0.9525) under the column-dropping strategy. Models excluding vitamin D maintained comparable accuracy, suggesting minimal impact on overall predictive performance. Conclusions: Machine learning models demonstrated high accuracy and robustness in predicting UTI subtypes across preprocessing strategies. While vitamin D contributes as a supportive feature, it is not essential for reliable prediction. These findings highlight the adaptability and clinical utility of both vitamin D-inclusive and vitamin D-exclusive models, supporting deployment in diverse healthcare settings. Full article

Accurate spleen segmentation in abdominal CT scans remains a critical challenge in medical image analysis due to variable morphology, low tissue contrast, and proximity to similar anatomical structures. This paper presents an enhanced U-Net architecture that addresses these challenges through multi-slice contextual integration and interpretable deep learning. Our approach incorporates three-channel inputs from adjacent CT slices, implements a hybrid loss function combining Dice and binary cross-entropy terms, and integrates Grad-CAM visualization for enhanced model interpretability. Comprehensive evaluation on the Medical Decathlon dataset demonstrates superior performance, with a Dice similarity coefficient of 0.923 ± 0.04, outperforming standard 2D approaches by 3.2%. The model exhibits robust performance across varying slice thicknesses, contrast phases, and pathological conditions. Grad-CAM analysis reveals focused attention on spleen–tissue interfaces and internal vascular structures, providing clinical insight into model decision-making. The system demonstrates practical applicability for automated splenic volumetry, trauma assessment, and surgical planning, with processing times suitable for clinical workflow integration. Full article