Jul 2026· Jurnal Ilmu Komputer dan Informasi· Vol 19, pp. 157-167· 0 citations· 20 references
TL;DR
The proposed framework establishes a reliable and lightweight baseline for automated gastrointestinal disease detection and demonstrates that ConvNeXt-Tiny effectively captures disease-relevant visual patterns in endoscopic images while maintaining consistent performance across varying training conditions.
Abstract
Early and accurate detection of gastrointestinal abnormalities, such as gastroesophageal reflux disease (GERD) and intestinal polyps, is essential for preventing severe clinical complications. However, manual interpretation of endoscopic images is often constrained by inter-observer variability and time limitations. This study proposes a ConvNeXt-Tiny-based deep learning framework for multi-class classification of gastrointestinal endoscopic images. Experiments were conducted using the GastroEndoNet v3 dataset, which contains 4,006 images categorized into four classes: GERD, GERD Normal, Polyp, and Polyp Normal. A total of twelve experimental scenarios were designed to systematically evaluate the effects of dataset-provided augmentation, ImageNet-based normalization, and batch size on model performance. The optimal configuration, combining augmentation, normalization, and a batch size of 64, achieved a test accuracy of 99.75% and a macro-averaged F1- score of 0.9977, indicating stable convergence and strong generalization on unseen data. The results demonstrate that ConvNeXt-Tiny effectively captures disease-relevant visual patterns in endoscopic images while maintaining consistent performance across varying training conditions. Comparative evaluation with a transformer-based baseline further indicates that modern convolutional architectures remain competitive for gastrointestinal image classification tasks. The proposed framework establishes a reliable and lightweight baseline for automated gastrointestinal disease detection. Extensions to video-based endoscopy would require incorporating temporal information across consecutive frames, which is beyond the scope of the current image-based study.
Intestinal diseases such as polyps, esophagitis, and ulcerative colitis may lead to serious complications when they are not detected early. Endoscopy provides important visual information for diagnosis; however, manual interpretation still depends on clinical expertise and requires time. This study aimed to develop an automatic intestinal disease classification model based on Convolutional Neural Network (CNN). The study used a Research and Development method consisting of data collection, preprocessing, model training, evaluation, and web interface implementation. The dataset was obtained from Kaggle and consisted of 6,000 endoscopic images categorized into normal, polyps, esophagitis, and ulcerative colitis. Each class contained 1,500 images, divided into approximately 87% training data and 13% testing data. The model was trained for 15 epochs and evaluated using accuracy, loss, confusion matrix, and single-image testing. The results showed stable validation accuracy in the range of 98-99%, while single-image testing produced confidence scores from 99.91% to 100%. This system is recommended as an initial endoscopic image classification aid, with further development involving a random-image class and an examination history database.
Gunawan, Muhtar, Lili Ruhyana et al.· Jurnal Teknologi Informatika...· 0 citations
This study evaluates the use of deep learning methods for multi-class classification and polyp segmentation on Kvasir endoscopic images. A dual-model approach was employed, where EfficientNet handles multi-class classification and U-Net handles polyp segmentation, each trained and evaluated independently on their respective datasets. The EfficientNet-B0 model achieved high performance, with accuracy, precision, recall, and F1-score values exceeding 91%, demonstrating its effectiveness in detecting various gastrointestinal abnormalities across eight classes. The U-Net model, while showing strong performance in background detection, faced challenges in lesion delineation, achieving a Dice Similarity Coefficient (DSC) of 34.50% and IoU of 20.85%. These results suggest that running both models in parallel on the same input image could provide simultaneous classification and segmentation outputs, offering more comprehensive diagnostic information compared to single-task approaches. This study contributes to the understanding of independent deep learning components that could support AI-based medical decision-making in gastrointestinal endoscopy.
Rahman Ardi Saputra, Suhendro Yusuf Irianto, Egi Safitri· Jurnal Nasional Pendidikan T...· 0 citations
Introduction Proper interpretation of the colonoscopic images is important to early detect and diagnose colorectal diseases like polyps and inflammatory bowel diseases. However, the complex visual patterns and high intra-class similarity of such images make the automated classification task challenging. Methods In this work, we propose an attention enhanced deep learning framework using ConvNeXt V2 for robust multi-class classification of colonoscopic images. The proposed method employs a Convolutional Block Attention Module (CBAM) in ConvNeXt V2 architecture to improve the feature representation by emphasizing the diagnostically relevant regions and ignoring the irrelevant background information. We used a balanced dataset of three classes: cecum (normal), polyp and ulcerative colitis with a uniform spatial resolution of 720 × 576 pixels. To improve the generalization of the model, we performed data augmentation for the training. The performance of the proposed model was extensively evaluated using 5-fold stratified cross-validation. Results Experimental results show that the proposed approach achieves a mean classification accuracy of about 95% which is significantly better than the baseline ConvNeXt V2 model which achieved about 90% accuracy. Furthermore, the proposed model achieved a mean precision of 95.1% and an F1-score of 94.9%, which shows a reliable classification of all classes. Moreover, qualitative analysis by attention visualization reveals that the model can focus on clinically relevant areas related to pathological features. Discussion The results demonstrated the effectiveness of modern convolutional architectures with embedded attention mechanisms in improving diagnostic performance in the analysis of colonoscopic images. The proposed framework provides a powerful and efficient tool for automatic classification of colorectal diseases and can assist clinicians for decision making.
Xiaosheng Jin, Lu-Xi Chen, Liwei Xue et al.· Frontiers in Oncology· 0 citations
Background: Deep-learning systems for endoscopic image classification are commonly evaluated with random data splits, which may overestimate performance under acquisition shift; uncertainty-aware selective prediction may improve reliability by allowing a model to abstain on uncertain cases. Methods: We evaluated binary abnormality detection in two endoscopic imaging domains: laryngeal contact-endoscopy narrow-band imaging (CE-NBI; 210 patients, patient-level) and gastrointestinal endoscopy (HyperKvasir; 6746 images, image-level, predominantly white-light). ImageNet-pretrained ResNet-50 deep ensembles were assessed under a resolution-defined acquisition-shift stress test; for the gastrointestinal data, a random stratified split was additionally used as an in-distribution reference. We evaluated discrimination, calibration, decision-curve analysis, and entropy-based selective prediction. Results: In the gastrointestinal dataset, random-split evaluation produced high performance (AUROC 0.994, 95% CI 0.990–0.996; AUPRC 0.991). Under resolution shift on the same data, performance fell to AUROC 0.723 (95% CI 0.710–0.736; AUPRC 0.690; sensitivity 0.417; specificity 0.874); the two intervals do not overlap. Selective prediction improved reliability among retained cases: under resolution shift, accuracy rose from 0.683 at full coverage to 0.852 (95% CI 0.835–0.871) at 25% coverage (balanced accuracy 0.646 → 0.798). Predictive entropy was significantly higher for incorrect than for correct predictions in both regimes (Mann–Whitney p = 7.1 × 10−77 with rank-biserial |r| = 0.30 under shift). In the laryngeal cohort, no statistically significant differences were detected among four architectures (ROC-AUC 0.844–0.901; all pairwise DeLong p > 0.05). Conclusions: Random-split evaluation substantially overestimated performance relative to a resolution-defined acquisition-shift stress test, and entropy-based selective prediction improved reliability by identifying a high-confidence subset for automated prediction while deferring the remainder to human review. Target-domain recalibration substantially restores calibration under shift (ECE 0.172 → 0.036 with temperature scaling; → 0.017 with isotonic regression) but does not recover discrimination; selective prediction is complementary, mitigating residual confident-wrong predictions. An encoder-transfer experiment showed asymmetric cross-domain utility; features learned on the larger gastrointestinal cohort transferred to the laryngeal cohort (AUROC 0.80 vs. in-domain 0.89), whereas the reverse direction did not transfer (0.53 vs. 0.72). Prospective multi-center validation remains required before clinical deployment. All code, fold definitions, random seeds, and a reproducible protocol are publicly released.
Behnam Kiani Kalejahi, S. Khan, Murodbek Akhrorov et al.· Biomedicines· 0 citations
Gastrointestinal (GI) tract diseases occur due to abnormalities affecting different regions of the digestive system. Diagnosing GI tract diseases typically involves methods such as endoscopy, imaging examinations, biopsy analysis, and clinical assessments. The severity and stage of the disease play a crucial role in determining the appropriate treatment plan. Medical imaging techniques are used to identify abnormal regions and assess the progression of diseases. The categorization of GI tract diseases, including ulcerative lesions, polyps, bleeding regions, inflammatory conditions, and normal tissues, is crucial for accurate diagnosis and effective treatment planning. Recently, Artificial Intelligence (AI)-based Deep Learning (DL) models have received greater attention due to their accuracy and flexibility. In this work, a threefold DL model is suggested to classify GI tract disease types. An improved ResNet, called Channel-Spatial Context Gating ResNet (CSCG-ResNet), is proposed for efficient feature extraction. Then, a modified Mountain Gazelle Optimizer (MGO) is proposed for feature optimization. Finally, the Quantum-inspired TabNet is proposed for multi-class categorization. The performance of the model is validated using the Kvasir dataset. The model achieves an overall accuracy of 96.75% when compared to previously proposed models.
Harshitha E.S., N. G., S. B et al.· Journal of Innovative Image...· 0 citations
Ulcerative Colitis Detection and Severity Prediction research creates an effective and dependable automated model to identify the severity of colon diseases using Wireless Capsule Endoscopy (WCE) images. Current methods typically use single deep learning models or conventional machine learning models, which do not readily model both fine-grained variations in mucosal texture and global contextual interactions, particularly when applied to small medical data sets. The improvements of generalization were done by data augmentation and training of the model (categorical cross-entropy loss) with optimized hyperparameters. It was applied to the Python platform with the deep learning libraries and tested on the WCE Curated Colon Disease Dataset, comprising 800 images and four severity levels. The suggested method had a precision of 97.5, which is a better performance than the current models. This system has advantages because it offers accurate diagnosis with computer-assisted assistance to gastroenterologists and aids in the early and accurate evaluation of the severity of UC. The proposed model uniquely combines- Local convolutional features via ResNet-50, Global contextual features via Vision Transformer, and Handcrafted clinical texture descriptors (GLCM + LBP). This multi-source feature fusion is reduced via PCA to preserve 95% variance and addresses the core limitations of single-architecture models that tend to either underfit local texture patterns or miss long-range spatial dependencies.
S. Dhole, C. S. More, Anuradha S. Nigade et al.· International Journal of Ele...· 0 citations