International Journal of Bioinformatics Research and Applications (15 papers in press)

Regular Issues

• A Comparative Study of Object Detection Techniques for Identifying Biomedical Waste  
  by Dinesh Deshmukh, Somnath Bhattacharya, Shubhashis Sanyal 
  Abstract: This paper focuses on methods for detecting biomedical waste obtained from healthcare facilitiest. Biomedical waste should be collected according to categories, rules, and regulations. Biomedical waste is more harmful and infectious compared to normal waste; therefore, it should be collected by robots to prevent infection. In this study, a comparison-based study is presented on the results obtained using the MobileNet, InceptionNet, and EfficientNet algorithms for object detection purposes. Approximately 1200 images of biomedical waste were used for training, testing, and validation purposes..These images of waste were processed through different models. Three models were compared in terms of their accuracy, model losses, prediction level, and unpredicted objects. Thus, according to the results, the MobileNet model outperformed the other two models. The MobileNet model can be used to identify biomedical waste, and people can have healthier lives by avoiding infections. So they can perform with high productivity and efficiency.
  Keywords: Object detection; biomedical wastes; classification; MobileNet ; InceptionNet; EfficientNet.
  DOI: 10.1504/IJBRA.2026.10073418
   
  
• Optimised Skin Disease Prediction Using Enhanced CRNN and Buzzard Optimisation in Deep Learning  
  by D. Ushanandini, T. Kamalakannan 
  Abstract: Deep Learning (DL)-based predictive dermatology models aim to reduce the global skin disease burden. Traditional systems might be faster or more accurate. Different DL methods classify skin diseases effectively. DL, recently gaining popularity, excels in complex domains like biomedicine with sparse clinical evidence. DL models extract features well and learn complex patterns accurately. Traditional systems are slow, and early Machine Learning (ML) strategies have shortcomings. This paper introduces the BUZOA using an ECRNN. Initially, the Melanoma Skin Cancer dataset from Kaggle was used. Image entities were pre-processed with a filter method to normalise pixels and minimise noise. Sliding segmentation in the ANFIS accurately masked boundary regions. Cancer sites were detected by colourimetric normalisation of slide sections. Features' weights were scaled by Individual Colour Histogram Equalisation (ICHE). The BUZOA method selects a smaller feature vector, reducing its length. ECRNN uses a softmax activation function to cluster BUZOA image features. Accurate feature sizing leads to better classification and prediction of skin diseases. The experimental dataset had 10,000 images; the proposed method outperformed SVM, CNN, and SVR in classification.
  Keywords: Skin Diseases; Enhanced Convolutional Recursive Neuron Network (ECRNN); Enhanced Convolutional Recursive Neuron Network; Colour Histogram Values; Adaptive Neural Fuzzy Inference System (ANFIS);.
  DOI: 10.1504/IJBRA.2026.10073778
   
  
• Versatile DeepFusion Classifier (VFC) for COVID-19 Detection in X-Ray and CT Images with Optimised RCNN-Based Segmentation  
  by Mairembam Stelin Singh  
  Abstract: This work details a COVID-19 detection model in five key phases, starting with data collection from CT and X-ray images. Regions of Interest (ROI) are pinpointed within the pre-processed images using an innovative and optimized recurrent convolution neural network O-RCNN model. Various features, encompassing color, texture, and shape, are carefully extracted from segmented data. Then, to identify the most informative features, a hybrid optimization model Golden Jackal Herd Dynamics with Status Vector (GJHDSV) is employed. This model combines Golden Jackal Optimization (GJO) and Coronavirus Herd Immunity Optimizer (CHIO). Finally, the selected optimal features are employed to train the Versatile DeepFusion Classifier (VFC), a multi-modal classifier. The VFC harnesses a spectrum of advanced architectures and techniques, including InceptionResNetV2, Quantum Neural Network (QNN), Long Short-Term Memory (LSTM) with an attention mechanism. The model attained impressive accuracy, reaching 99.12% for X-ray and 98.39% for CT classification.
  Keywords: Covid 19; O-RCNN; GJO; CHIO; VFC; GJHDSV.
  DOI: 10.1504/IJBRA.2026.10074636
   
  
• Enhancing HIV/AIDS Prediction Models: a Comparative Analysis of Machine Learning Algorithm  
  by Shivam Tiwari, Surbhi Vijh 
  Abstract: The global HIV/AIDS epidemic remains a significant public health challenge, necessitating robust and accurate predictive models for effective disease management. Despite advancements in machine learning algorithms, achieving high predictive accuracy in HIV/AIDS classification tasks remains elusive. In this study, the gap is addressed by evaluating various machine learning algorithms, including SVM, Naive Bayes, Random Forest, AdaBoost, Decision Tree, K-Nearest Neighbours, Logistic Regression, and XG Boost, alongside Gradient boosting both with default parameters and hyperparameter tuning. The comprehensive dataset is employed and rigorous evaluation metrics to assess the performance of each algorithm. The extensive results indicate that Gradient Boosting with hyperparameter tuning achieves the highest accuracy of 70.82%, with notable precision, recall, and AUC-ROC scores. These findings underscore the potential of advanced machine learning techniques in enhancing HIV/AIDS prediction models. By optimising model parameters and leveraging ensemble learning strategies, our approach demonstrates promising outcomes in AIDS infection analysis.
  Keywords: HIV/AIDS; Machine Learning; Classification; Opportunistic infections; CD4 cells.
  DOI: 10.1504/IJBRA.2026.10074996
   
  
• A Systematic Review: Investigating Trastuzumab as A Targeted Therapeutic Treatment for Inhibiting HER2-Positive Gastric Cancer, utilising Phytomolecules Derived from Aegle Marmelos  
  by Md. Tofazzal Hosen, Ibrahim Al Imran 
  Abstract: Worldwide, stomach cancer is a common cancer with a prognosis of less than a year when it is incurable or has spread to other areas of the body. Our comprehension of gastric cancer has greatly advanced in recent years due to the substantial amount of HER2-related information gathered from studies on stomach cancer. This has led to a clearer understanding of the incidence of HER2 amplification in gastric cancer. HER2 expression stands out as the primary biomarker guiding the incorporation of trastuzumab into first-line systemic chemotherapy, ultimately enhancing overall survival in advanced HER2-positive gastric carcinoma. In HER2-amplified gastric cancer, inhibiting Human epidermal growth factor receptor 2 (HER2) is a vital therapeutic approach. Furthermore, HER2-targeted treatments have shown encouraging outcomes in the context of gastric cancer with HER2 positivity. It's worth highlighting that the field of HER2-targeted therapies is swiftly evolving, incorporating a variety of approaches such as small-molecule inhibitors, antibody-drug conjugates, and bispecific antibodies. This review presents a compilation of phytocompounds with the potential for repurposing targeted therapeutic approaches or medications specifically targeting HER2-positive cancer cells.
  Keywords: Gastric cancer; Human epidermal growth factor receptor 2; Trastuzumab; Biomarkers; antibody-drug conjugates; and bispecific antibodies.
  DOI: 10.1504/IJBRA.2026.10075173
   
  
• Review on ECG Encryption Methods  
  by Fatma Zohra Besmi, Samia Belkacem, Noureddine Messaoudi 
  Abstract: With the growing reliance on digital technology, protecting sensitive data like electrocardiography (ECG) is more crucial than ever. ECG encryption techniques are pivotal for ensuring data privacy. This research reviews recent advancements in this area, focusing on four main techniques: elliptic curve cryptography (ECC), homomorphic encryption (HE), the Advanced Encryption Standard (AES), and chaotic encryption. The study delves into each of these techniques in detail, focusing on their strengths and weaknesses as well as evaluating their performance through a set of metrics such as positive prediction rate (+P%), sensitivity (Se%), error detection rate (DER%), accuracy (Acc%), and data compression ratio (CR%). This study offers valuable insights into the current ECG encryption landscape. It guides future research towards the development of secure and efficient solutions for safeguarding sensitive cardiac data and enhancing patient care quality in the digital age.
  Keywords: AES; chaotic encryption; data privacy; ECG signal; elliptic curve encryption; encryption; homomorphic encryption.
  DOI: 10.1504/IJBRA.2026.10075229
   
  
• Identification of Individuals by Biometric Fusion of the Palmprint and the Ear  
  by Chahreddine Medjahed, Narimane Wafaa Krolkral 
  Abstract: This work proposes a biometric multimodal system based on feature fusion extracted from the ear and palmprint. The first step consists of developing monomodal models based on Convolutional Neural Networks (CNNs) using the AMI dataset for the ear and the IITD dataset for the palmprint. These models were benchmarked for extracting the best-performing features to fuse through concatenation. The second step consists of using machine learning for classification and recognition. The effectiveness of our approach was tested under noisy conditions (salt-and-pepper and Gaussian noise), demonstrating robust performance.
  Keywords: Biometric; Multimodal system; Feature fusion; Convolutional Neural Networks; Machine learning.
  DOI: 10.1504/IJBRA.2026.10075316
   
  
• Papilledema Detection using DenseNet-201 Architecture and Grad-CAM  
  by Abdelhak Mehadjbia, Fouad Slaoui-Hasnaoui 
  Abstract: Early detection of retinal abnormalities is vital, particularly for severe and complicated cases. Availability of retinal fundus imaging and sophisticated CNN frameworks that have been extensively utilized in the medical sector with demonstrated outcomes. Consequently, a system for detecting Papilledema can enhance diagnostic quality, save time, and offer doctors extensive information about the patient, particularly when conveying this detection to an ophthalmologist to expedite the treatment process. Therefore, our method utilises the DenseNet201 architecture with a transfer learning technique to identify Papilledema among six additional classes, such as, Pseudo-papilledema, Glaucoma, Cataract, Age-related macular degeneration (ARMD), Diabetic Retinopathy (DR), and Healthy retina. This experiment is conduct with 10610 images from 3 different dataset sources. The proposed model achieved an accuracy of 96.1%, precision of 96.3%, recall of 95.8%, and F1-score of 96.3%. In addition, we have used Grad-CAM, which introduces model prediction's interpretations in our study.
  Keywords: Retinal disease; Papilledema; Artificial intelligence; Deep neural network; Model interpretation; Model explainability; Grad-Cam.
  DOI: 10.1504/IJBRA.2026.10076526
   
  
• A RAG-Based Approach to Diabetic Foot Ulcer Knowledge Retrieval and Question Answering  
  by Bhavika Patel, Vibha Patel 
  Abstract: Diabetic foot ulcers are a severe and common complication of diabetes, often leading to infection and lower limb amputation. Early and effective treatment is crucial. The internet offers millions of documents to support effective treatment strategies for this condition. However, accurate information extraction from high-dimensional datasets is still quite difficult. Traditional Information Retrieval approaches rely on keyword matching, statical ranking and rule-based approaches to search the relevant documents. These approaches struggle to retrieve relevant knowledge before generating the well-formed answers. To overcome these limitations, this research highlights the potential of AI-driven technologies like Large Language Model, Retrieval Augmented Generation and FAISS vector database. This hybrid approach enables structured document processing, context-aware text chunking, and interactive communication via a web interface by extracting semantics from multiple PDFs. The study aims to bridge the gap between healthcare professionals and the knowledge in documents, promoting LLM-driven solutions to improve decision-making.
  Keywords: Context-aware semantics; Diabetic foot ulceration; Information Retrieval; Large Language Model; Retrieval-Augmented Generation.
  DOI: 10.1504/IJBRA.2026.10076938
   
  
• Artificial intelligence and machine learning in temporomandibular disorders: a comprehensive synthesis of current review and meta-analysis evidence  
  by Neada Hysenaj, Medjon Hysenaj, Vergjini Mulo 
  Abstract: Artificial intelligence (AI) and machine learning (ML) are revolutionising diagnostic processes in the field of dentistry. This study aims to summarise evidence from reviews and meta-analyses on AI-based diagnosis, predictive modelling, and educational applications related to the temporomandibular disorders. Sixteen reviews were identified through PubMed, Scopus, and Web of Science that were related to AI and ML in the TMJ field. Deep learning (DL) models achieved higher diagnostic precision for the diagnosis of TMJ osteoarthritis and internal derangements. Ensemble ML models showed good accuracy with easy interpretation. MRI-based AI systems showed high accuracy in determining the disc displacement, while neuroimaging models showed changes in brain areas linked to pain. AI and ML show considerable potential for improving the accuracy and consistency of TMD diagnosis. However, heterogeneity across reviews and the lack of standardised quality assessment limit clinical translation.
  Keywords: artificial intelligence; machine learning; temporomandibular joint; deep learning; radiomics; predictive modelling; diagnostic imaging.
  DOI: 10.1504/IJBRA.2026.10078210
   
  
• Enhancing Diagnostic Precision of Early Stage Squamous Cell Carcinomas through Hybrid Feature Selection and Diverse Classifiers: An Evaluation in Oral Cancer Prediction  
  by Tamilselvi K, Rekha V, Hemalatha P, Manikandan J, Jayashree K 
  Abstract: This study investigates the complex task of accurately predicting whether primary stage squamous-cell carcinoma will present itself or not, using an initial set of 25 features to predict various stages of oral carcinoma. This research utilizes a hybrid feature selection technique to reduce 25 initial qualities to 14 manageable features, thereby decreasing both computational complexity and improving predictability of its model. Four different classifiers were employed to implement the process of prediction. Data is transformed by using Synthetic Majority Over-sampling Technique and this could improve prediction model performance. A notable finding was made: support vector machines emerged as the premier machine learning technique after SMOTE integration displaying superior performance. This study emphasizes the significance of accuracy when diagnosing disease, as well as feature selection techniques tailored towards increasing predictive precision. A hybrid feature selection technique, a unified system of medical information.
  Keywords: Linear Regression (LR); Logistic regression (LOR); Naïve bayes (NB); Random Forest (RF) and Artificial Neural Network (ANN).
  DOI: 10.1504/IJBRA.2026.10078499
   
  
• Discovery of Putative LpxC Inhibitors against Klebsiella pneumoniae: An Early-Stage Rationale Structure-Based Drug Design  
  by Manish Bhadana, Rashmi Prabha Singh, Monika Jain, Jayaraman Muthukumaran, Amit Kumar Singh 
  Abstract: Klebsiella pneumoniae, a major ESKAPE pathogen, poses a serious global health risk, causing pneumonia, urinary tract, and bloodstream infections, especially in newborns, the elderly, and immune compromised patients. The rise of multi-drug-resistant (MDR) strains has led to mortality rates up to 70% in hospitals. Targeting the lipid A biosynthesis pathway, an essential but underexplored antibacterial target in Gram-negative bacteria, this study identified the zinc-dependent enzyme LpxC as a promising therapeutic focus. Using an optimised LpxC model from K. pneumoniae strain 342, over 4.4 million compounds from the Enamine database were screened virtually. Three compounds Z2091574139, Z4002426309, and Z3888823409 showed strong binding and favourable pharmacological profiles. Molecular dynamics and MM/PBSA analyses identified Z2091574139 (41.74 kcal/mol) as the top candidate. These findings suggest Z2091574139 as a potential lead for novel anti-K. pneumoniae agents to combat MDR infections.
  Keywords: ESKAPE; LpxC; K.pneumoniae; Virtual screening; ADME; MD Simulation; MM/PBSA and HTVS.
  DOI: 10.1504/IJBRA.2027.10079644
   
  
• Data Augmentation in Healthcare Data Modalities: A Comprehensive Review  
  by Sarmistha Nanda, Soudaminee Sahoo, Chhabi Rani Panigrahi, Bibudhendu Pati 
  Abstract: Data Augmentation (DA) strategies offer a solution by synthesizing the existing data to improve model performance. In the healthcare domain, obtaining well-annotated datasets remains challenging. Therefore, integrating DA approaches have become essential for developing medical applications. The review focuses on studies conducted between 2020 and 2025 that includes multiple data modalities and their integration with machine learning and deep learning models. This study considers a detailed review on DA techniques across various data modalities such as medical imaging, clinical text, structured tabular data, biomedical video, and physiological audio signal data for review. It further explores the methodological variations of DA techniques across these modalities and the associated benefits as well as challenges. The review findings indicate that the effectiveness of DA techniques often depends on the data modality and the underlying model architecture. Furthermore, this review examines diverse validation techniques used in the literature to evaluate generated synthetic data.
  Keywords: Data Augmentation; Healthcare; Machine Learning; Deep Learning; Medical Imaging; Generative Models; Synthetic Data; Medical Data Modalities.
  DOI: 10.1504/IJBRA.2027.10079918
   
  
• ChromoFormer: Integrating Artificial Intelligence and Cytogenetics for high throughput chromosome analysis  
  by DINU A. G, Biju V. G, Vinod B. R, Nonam Chellapan 
  Abstract: Chromosome segmentation is critical in cytogenetics, aiding diagnosis of genetic disorders and cancer. This paper introduces ChromoFormer, a two-stage method combining image enhancement with a modified vision transformer for accurate chromosome segmentation. Stage one enhances metaphase spread image quality; stage two performs precise segmentation using the transformer architecture. The model was validated on two large datasets: the Cell Image Library (CIL), comprising 5,000 metaphase images across 23 chromosome pairs with 229,852 annotations (2,000 per chromosome type) and 5,000 pixel-level segmentation labels; and the Bio Image Chromosome Classification Dataset (BICCD), containing 42 chromosome objects with associated weights and annotation files. On the CIL dataset, ChromoFormer achieved 97.8% overall accuracy, 96.5% precision, and a Cohens Kappa of 0.962, indicating excellent inter-rater agreement. Segmentation quality was confirmed by a Jaccard Index of 0.934 and Dice coefficient of 0.951. Image reconstruction fidelity was demonstrated via MSE of 0.0023, PSNR of 38.2 dB, and SSIM of 0.985 establishing ChromoFormers state-of-the-art performance
  Keywords: Deep learning; Segmentation; Hybrid model; Chromosome image.
  DOI: 10.1504/IJBRA.2026.10080454
   
  
• In Silico Analysis of Neem-Derived Proteins as Inhibitors of Key Antimicrobial Resistance Determinants in Gram-Positive and Gram-Negative Bacteria  
  by Mohammed Al Saiqali, Ayla Sanjay, Chand Pasha 
  Abstract: The swift rise of multidrug-resistant (MDR) bacterial infections necessitates novel antimicrobials. Neem peptides from Azadirachta indica possess broad medicinal properties. In this study, neem leaf proteins were isolated, analysed by SDS-PAGE, sequenced via MALDI TOF/TOF, and its 3D models generated. Five validated neem-derived proteins CytochromeP450, Squalene Epoxidase1, Cytochromef and Putative LOV Domain-Containing Protein were docked against -lactamase, penicillin-binding protein 2a, and the AcrB efflux transporter. Ramachandran plots, clash score evaluation, and rotamer outlier detection were performed by Procheck. The highest interactions (30) between aminoacids were observed in the case of squalene epoxidase 1 neem protein with the target penicillin binding protein 2a. Cytochrome P450 exhibited the strongest docking across all targets, particularly with penicillin binding protein 2a having force field energy 93,308.906. Most of the neem derived proteins were showing strong interactions with penicillin binding protein 2a and -lactamase than AcrB efflux compared to control Potassium clavulanate, with -lactamase.
  Keywords: Insilico; Azadirachta indica; antimicrobial resistance; Ramachandran plots; Structural Modeling.
  DOI: 10.1504/IJBRA.2027.10080645