WoS İndeksli Yayınlar Koleksiyonu

Permanent URI for this collectionhttps://hdl.handle.net/20.500.14627/6

Browse

Filters

20262

Settings

Author: search.filters.author.Pacal, IshakPublication Index: search.filters.publicationIndex.WoS

Search Results

Now showing 1 - 2 of 2
  • Article
    DiSCNet: Directional Split Convolution for Compute-Efficient Brain Tumor Diagnosis
    (Elsevier Sci Ltd, 2026) Pacal, Ishak; Ganie, Shahid Mohammad
    Brain tumor classification from magnetic resonance imaging (MRI) remains challenging because tumor appearance varies substantially across patients, while scanner- and protocol-related differences can alter image intensity distributions and weaken model generalization. This study aims to develop a compact yet highperforming deep learning framework that can classify heterogeneous brain MRI more reliably without relying on large-capacity backbones. To this end, we propose DiSCNet, a lightweight architecture built on an InceptionNeXt-inspired hierarchical design and centered on a novel Directional Split Convolution (DiSC) block. The proposed block diversifies receptive fields through complementary local and directional depthwise branches, while Global Response Normalization and Efficient Channel Attention are incorporated to improve feature stability and channel selectivity under acquisition variability. The model was evaluated on a unified benchmark constructed from five publicly available MRI repositories, comprising 17,888 images across four classes: glioma, meningioma, pituitary and non-tumor. Under a single training and evaluation protocol, DiSCNet was compared against 71 contemporary convolutional, transformer-based, and hybrid architectures. DiSCNet achieved the best overall performance, with 0.9922 accuracy, 0.9916 precision, 0.9930 recall, and 0.9923 F1-score, while using only 2.78 million parameters. Class-wise analysis further showed strong and balanced recognition across all diagnostic categories, and Grad-CAM visualizations indicated that the model predominantly focused on lesionrelevant regions. These findings demonstrate that a carefully designed lightweight architecture can outperform substantially larger models and provide an efficient, robust, and clinically relevant solution for four-class brain tumor MRI classification.
  • Article
    Denta-Hybridonet: A Hybrid CNN-Transformer Architecture for Automated Detection of Developmental Dental Anomalies in Pediatric Panoramic Radiographs
    (Elsevier Sci Ltd, 2026) Eskibaglar, Busra Karaagac; Yavuz, Yelda Polat; Dogan, Gizem Karagoz; Algarni, Ali; Cakmak, Yigitcan; Pacal, Ishak
    Accurate identification of developmental dental anomalies (DDAs) in children is clinically important; however, interpreting panoramic radiographs can still vary across readers because of mixed dentition, anatomical overlap, and variable image quality. This variability may delay recognition and complicate early interventional planning. In this study, we curated a pediatric panoramic dataset of 2,001 radiographs (ages 6-14 years) spanning five categories: Dilaceration, Ectopy, Hypodontia, Taurodontism, and Healthy. All images were independently labeled by three experienced pediatric dentists. To avoid patient-level leakage, the dataset was divided into training, validation, and held-out test sets using a patient-wise split. We propose Denta-HybridoNet, a hybrid convolution-transformer architecture designed to capture both fine-grained tooth morphology and broader, arch-wide contextual patterns. Its InceptionNeXt-gMLP block supports multi-scale local representation learning, which helps the model focus on subtle morphological cues, whereas the Swin-gMLP block provides efficient global context modeling across the dental arch. In addition, a gated multilayer perceptron (gMLP) module refines the feature transformation through context-dependent modulation, strengthening diagnostically relevant signals while reducing the influence of irrelevant variation and radiographic noise. To ensure a fair comparison, we benchmarked Denta-HybridoNet against 22 recent convolutional and transformer-based models under the same training protocol and evaluation conditions. On the held-out test set, the proposed method achieved 91.15% accuracy and 91.20% F1 score, representing the best overall performance among the compared architectures. Ablation studies quantified the contributions of hybrid design and gMLP, and Grad-CAM analyses supported interpretability by highlighting clinically meaningful regions.