Jul 2026· Jurnal Nasional Pendidikan Teknik Informatika (JANAPATI)· 0 citations
TL;DR
The 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.
Abstract
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.
This study integrates segmentation, transfer learning, and clinical validation to present a lightweight deep learning architecture for endoscopic lesion categorization. The Bionnica Lite architecture, a small convolutional neural network intended to provide a good classification performance with less computing cost, is at the center of the strategy. A segmentation module based on SAM 2.1 is included to improve lesion-focused analysis, allowing for accurate region-of-interest identification and better feature representation. Using pre-trained encoders such as EfficientNet-B0, the system supports both direct and transfer learning. According to experimental data, Bionnica Lite dramatically reduces latency and parameter count while achieving competitive accuracy when compared to more sophisticated models. Unlike existing studies that primarily emphasize predictive accuracy, the proposed framework investigates the balance between diagnostic performance, computational efficiency, lesion localization, and clinical interpretability within a unified deployment-oriented pipeline.
Diogen Babuc, D. Onchis, Melania Ardelean· Algorithms· 0 citations
Background: Accurate segmentation and classification of multiple sclerosis (MS) lesions are vital for a reliable diagnosis and disease monitoring. However, lesion heterogeneity in size, location, and intensity poses significant challenges to automated analysis. Methods: To address this, we developed a dual-stage pipeline integrating deep learning (DL) for precise spatial delineation and machine learning (ML) for robust classification of MS lesions. Two advanced DL models, nnU-Net and UNETR++, were optimized for lesion segmentation. Moreover, UNETR++ and several conventional ML methods were considered for the classification task, and Random Forest was found to be the best choice. Results: Experimental results indicate that nnU-Net outperformed UNETR++ for lesion segmentation across all cases, achieving a maximum improvement of 12.8%. During classification, Random Forest consistently outperformed advanced DL models, achieving at least 12% higher performance. Under practical conditions, an optimized hybrid pipeline that integrates nnU-Net for precise segmentation with Random Forests for robust classification delivers the best overall performance. Furthermore, qualitative analysis indicates that some apparent false positives may correspond to lesions missed during annotation, highlighting potential limitations in ground truth labeling. Conclusions: Overall, the proposed pipeline effectively leverages the complementary strengths of DL and ML, offering a promising, accurate framework for automated MS lesion analysis with potential clinical utility.
Reza Naghne, Mahdiyeh Rahmani, Ali Kazemi et al.· Diagnostics· 0 citations
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.
Muhammad Faqih, O. Q. Aziz, Ajib Hanani· Jurnal Ilmu Komputer dan Inf...· 0 citations
BACKGROUND AND OBJECTIVE
Contrast-enhanced ultrasound (CEUS) is widely used for evaluating thyroid nodule malignancy, but conventional time-intensity curve (TIC) analysis is labor-intensive and operator-dependent. This study proposes LSTAC, an automated framework for nodule segmentation and TIC analysis in CEUS videos.
METHODS
LSTAC integrates an improved YOLOv5-based segmentation network with a peak intensity frame (PIF) extraction algorithm to enable automatic nodule localization, TIC generation, and PIF identification. The framework was trained using CEUS data from 623 patients collected across three hospitals and evaluated on both internal and external validation cohorts.
RESULTS
LSTAC achieved 3-10× higher efficiency than VueBox in PIF extraction while maintaining strong temporal accuracy (0.94, 0.77, 0.79) and structural similarity (SSIM: 0.80, 0.60, 0.67). In malignancy prediction based on PIF features, LSTAC outperformed VueBox in two of three validation sets, with AUCs of 0.8279 vs. 0.8226 and 0.8000 vs. 0.7000.
CONCLUSION
LSTAC provides an efficient and accurate solution for automated CEUS analysis, reducing manual workload and improving consistency in thyroid nodule assessment, with potential for clinical application.
Aoxiang Yang, Liuyue Li, Ruifan He et al.· Artificial Intelligence in M...· 0 citations
An attention enhanced deep learning framework using ConvNeXt V2 for robust multi-class classification of colonoscopic images that demonstrates the effectiveness of modern convolutional architectures with embedded attention mechanisms in improving diagnostic performance in the analysis of colonoscopic images.
Xiaosheng Jin, Lu-Xi Chen, Liwei Xue et al.· Frontiers in Oncology· 0 citations
This work presents a leakage-controlled deep-learning framework for breast cancer classification using the CBIS-DDSM mammography archive. The proposed pipeline combines patient-level data partitioning before augmentation, a two-stage transfer-learning strategy based on ResNet50, and an Inter-View Attention Fusion (IVAF) module for adaptive fusion of paired craniocaudal (CC) and mediolateral oblique (MLO) feature maps. IVAF was modeled as a light-weighted convolutional gating strategy added after the last ResNet50 convolutional layer in order to create a weighted spatial-channel representation from the paired mammography images. In terms of the performance of the model under CBIS-DDSM held-out testing protocol, the entire model scored an accuracy of 97.12%, sensitivity of 96.44%, specificity of 97.68%, and AUC-ROC of 0.9876 based on the test results obtained on 6,117 images of 222 different patients. The average accuracy obtained using 100 random seeds was found to be 97.11% ± 0.18%.
Magy Makram, Alber S. Aziz, Mary Monir Saeid et al.· Scientific Reports· 0 citations