Browsing by Author "Akdag, Mehmet"

Now showing 1 - 3 of 3

Automated Mucormycosis Diagnosis from Paranasal CT Using ResNet50 and ConvNeXt Small
(MDPI, 2025) Toprak, Serdar Ferit; Dedeoglu, Serkan; Kozan, Gunay; Ayral, Muhammed; Can, Sermin; Turk, Omer; Akdag, Mehmet
Purpose: Mucormycosis is a life-threatening fungal infection, where rapid diagnosis is critical. We developed a deep learning approach using paranasal computed tomography (CT) images to test whether mucormycosis can be detected automatically, potentially aiding or expediting the diagnostic process that traditionally relies on biopsy. Methods: In this retrospective study, 794 CT images (from patients with mucormycosis, nasal polyps, or normal findings) were analyzed. Images were resized and augmented for training. Two transfer learning models (ResNet50 and ConvNeXt Small) were fine-tuned to classify images into the three categories. We employed a 70/30 train-test split (with five-fold cross-validation) and evaluated performance using accuracy, precision, recall, F1-score, and confusion matrices. Results: The ConvNeXt Small model achieved 100% accuracy on the test set (precision/recall/F1-score = 1.00 for all classes), while ResNet50 achieved 99.16% accuracy (precision approximate to 0.99, recall approximate to 0.99). Cross-validation yielded consistent results (ConvNeXt accuracy similar to 99% across folds), indicating no overfitting. An ablation study confirmed the benefit of transfer learning, as training ConvNeXt from scratch led to lower accuracy (similar to 85%) Conclusions: Our findings demonstrate that deep learning models can accurately and non-invasively detect mucormycosis from CT scans, potentially flagging suspected cases for prompt treatment. These models could serve as rapid screening tools to complement standard diagnostic methods (histopathology), although we emphasize that they are adjuncts and not replacements for biopsy. Future work should validate these models on external datasets and investigate their integration into clinical workflows for earlier intervention in mucormycosis.
Is the Ensemble Machine Learning Model a Reliable Method for Detecting Neoplastic Infiltration of Thyroid Cartilage in Laryngeal Cancers
(MDPI, 2025) Can, Sermin; Turk, Omer; Ayral, Muhammed; Kozan, Gunay; Onur, Mehmet; Yagiz, Eyyup; Akdag, Mehmet
Background and Objectives: We aimed to apply the ensemble machine learning model to diagnose thyroid cartilage invasion detected in computer tomography (CT) images in laryngeal cancers and evaluate the diagnostic performance of the model. Materials and Methods: A total of 313 patients were divided into two groups: the cartilage invasion group and the no cartilage invasion group. At least four CT slices were randomly selected for each patient, resulting in a total of 1251 images used in the study. A total of 619 axial CT images from the no cartilage invasion group and 632 axial CT images from the cartilage invasion group were used in the study. We reviewed the CT images and histopathological diagnoses in all cases to determine the invasion positive- or negative-status as a ground truth. The ensemble model, comprising ResNet50 and MobileNet deep learning architectures, was applied to CT images. Results: The following were obtained by the ensemble model with the test dataset: area under the curve (AUC) 0.99, and accuracy 96.54%. This model demonstrates a very high level of performance in detecting thyroid cartilage invasion. Conclusions: The ensemble machine learning model is an effective method for detecting neoplastic infiltration of the thyroid cartilage. Moreover, it may be a valuable diagnostic tool for clinicians in assessing disease prognosis and determining appropriate treatment strategies in laryngeal cancers. In conclusion, this model could be integrated into future clinical practice in laryngology and head and neck surgery for the detection of cartilage neoplastic infiltration.
Role of Machine Learning Segmentation Method Based on CT Images in Preoperative Staging of Oral Cavity Cancer
(Springer, 2025) Can, Sermin; Succo, Giovanni; Coskun, Cengiz; Korkmaz, Mehmet Hakan; Akdag, Mehmet
ObjectiveThe article aims to demonstrate, using oral cavity SCC as an example, that machine learning can accurately predict the T and N staging of OSCC, using the conventional radiologist/ surgeon interpretation of the scan as the reference standard.Materials and methodsTwo datasets for tumor mass and nodal metastasis were used in this study. Each of the datasets consists of 179 Contrast-enhanced Computed Tomography images. A customized U-Net deep learning architecture was employed for the segmentation of tumor masses and nodal metastases. Comprehensive maps of the tumor mass and metastatic lymph nodes were generated. Following this mapping process, the dimensions of the identified lesions were measured and classified according to the Tumor and Lymph Node Metastasis classification system. The resulting classifications were then compared with those established by a radiologist to assess accuracy.ResultsThe performance metrics for tumor mass and metastasis segmentation were as follows: binary accuracy value of 98.81% and 99.58%, respectively. The accuracy values were 75.00% for tumor grade classification and 97.22% for nodal status classification.ConclusionWe emphasize that machine learning-based segmentation methods effectively predict tumor mapping and staging in oral cavity tumors, demonstrating correlation with surgeons/radiologists' assessments. As such, this model can be a diagnostic tool that supports clinicians in making informed therapeutic decisions. We foresee that, with the continuous evolution of technology, the segmentation model employed in our study will undergo significant advancements, ultimately facilitating three-dimensional tumor mapping in the near future.