Diagnosis of most ophthalmic conditions, such as diabetic retinopathy, generally relies on an effective analysis of retinal blood vessels. Techniques that depend solely on the visual observation of clinicians can be tedious and prone to numerous errors. In this article, we propose a semi-supervised automated approach for segmenting blood vessels in retinal color images. Our method effectively combines some classical filters with a Generalized Linear Model (GLM). We first apply the Curvelet Transform along with the Contrast-Limited Histogram Adaptive Equalization (CLAHE) technique to significantly enhance the contrast of vessels in the retinal image during the preprocessing phase. We then use Gabor transform to extract features from the enhanced image. For retinal vasculature identification, we use a GLM learning model with a simple link identity function. Binarization is then performed using an automatic optimal threshold based on the maximum Youden index. A morphological cleaning operation is applied to remove isolated or unwanted segments from the final segmented image. The proposed model is evaluated using statistical parameters on images from three publicly available databases. We achieve average accuracies of 0.9593, 0.9553 and 0.9643, with Receiver Operating Characteristic (ROC) analysis yielding Area Under Curve (AUC) values of 0.9722, 0.9682 and 0.9767 for the CHASE_DB1, STARE and DRIVE databases, respectively. Compared to some of the best results from similar approaches published recently, our results exceed their performance on several datasets.

Background: Stroke is a leading cause of mortality and long-term disability worldwide. Rapid and accurate diagnosis is critical for effective management. Although manual diagnosis remains tedious and prone to potential errors, most existing automated analysis tools rely on complex architectures that are challenging to deploy in resource-limited clinical settings. To address this gap, we propose a novel lightweight decision-support approach for the preliminary classification of non-contrast brain CT scans into three categories: ischemic stroke, hemorrhagic stroke, and normal cases. Using an anonymized dataset from the Bafoussam Regional Hospital Center, 16 textural features (first-order and gray-level co-occurrence matrices) were extracted from manually targeted circular regions of interest to balance discriminative capability with computational efficiency. Results: Evaluated using a lightweight Random Forest classifier under patient-level stratified Group K-fold cross-validation, the proposed model achieved a validated mean accuracy of 0.9928 (± 0.0101). Precision reached 1.00, while recall and F1-scores ranged between 0.99 and 1.00 across all classes within the validation setup. The consistency of these results across validation folds suggests that the extracted texture features provide distinct signatures for the targeted stroke subtypes within this cohort. Conclusion: These performances demonstrate the feasibility of this radiomics-based approach. As a proof of concept, our approach shows promise as a preliminary decision-support tool with potential for future clinical workflows. It is promising for resource-limited settings where access to expensive and high-computational neuroimaging software is restricted. Clinical trial number: Not applicable.

Purpose: Timely and precise classification of stroke types is critical for immediate clinical intervention. This study aims to propose a lightweight dual-domain pipeline optimized for three-class ROI-level computed tomography (CT) stroke classification, bridging the gap between automated analysis and clinician-led diagnosis in resource-constrained settings. Methods: We introduce a computationally efficient hybrid feature extraction method combining discrete wavelet transform (DWT) and gray-level co-occurrence matrix (GLCM), leveraging a synergistic multi-scale textural analysis. Images are preprocessed according to the image biomarker standardization initiative (IBSI) standards (using fixed 8-bin gray-level discretization). From localized 17 × 17pixel region of interest (ROI) patches, a highly compact 8-dimensional dual-domain feature vector is constructed. This vector is fed into a lightweight multilayer perceptron (MLP) for 3-class classification (Hemorrhagic vs. Ischemic vs. Normal), ensuring full feature-level explainability. Robustness is guaranteed using a strict patient-grouped 5-fold cross-validation protocol to prevent data leakage. Results: The model achieved a peak diagnostic accuracy of 99.17% using clinical data from the Regional Hospital Center of Bafoussam, and maintained a high performance of 98.00% on the external Teknofest-2021 dataset. A comprehensive performance report table and a multi-class confusion matrix confirm the model's precision in distinguishing between tissue types. Furthermore, multi-class calibration curves demonstrate high diagnostic reliability and probability alignment, providing clinicians with explainable, well-calibrated confidence scores for decision-making. This framework is supported by a prototype interactive ROI-based diagnostic tool running efficiently on standard CPU hardware. Conclusion: The integration of DWT and GLCM with MLP classification demonstrates exceptional robustness and high precision, surpassing several state-of-the-art methods in recent literature. Owing to its low computational footprint, this framework offers a straightforward implementation path into existing clinical workflows, providing an accessible and reliable CAD tool for stroke management in resource-constrained emergency settings.