Creative Commons source reading · Machine learning for performance
Deep learning–based automated detection of oral squamous cell carcinoma in histopathological images: a comparative study of five CNN architectures
Yunus Balel; Kaan Sağtaş; Fatih Teke; Mehmet Ali Kurt
Originally published in Odontology · 2026-06-15
From the article
Selected source text. JournalOS removes markup and shortens excerpts where indicated. The full article provides the complete context.
Abstract
Oral squamous cell carcinoma (OSCC) is the most common malignancy of the oral cavity, and early diagnosis plays a crucial role in improving patient prognosis and survival rates. Histopathological examination remains the gold standard for OSCC diagnosis; however, this process is time-consuming and highly dependent on expert interpretation. With the rapid development of digital pathology and artificial intelligence, deep learning–based approaches have emerged as promising tools to support automated diagnostic systems. In this study, five convolutional neural network (CNN) architectures—VGG16, ResNet50, InceptionV3, EfficientNetV2S, and ConvNeXt-Tiny—were comparatively evaluated for the automated classification of OSCC using histopathological images. An open-access OSCC dataset was utilized, and two experimental scenarios were created…
Results
In the dataset without augmentation, a total of 1224 histopathological sections were used, with 979 images for training, 122 images for validation, and 123 images for the test dataset. In the augmented setting, data augmentation was applied only to the training set, increasing the effective number of training samples, while the validation (n = 122) and test (n = 123) sets remained unchanged. As a result, the training set was expanded through augmentation, whereas validation and test sets preserved their original distributions for unbiased evaluation. The distribution of OSCC and normal tissue in the training, validation, and test datasets is shown in Fig. 2. The classification results of the models were calculated using Accuracy, Precision, Recall, Specificity, F1-Score, and ROC-AUC metrics and are presented in Table 2.…
Discussion
This study provides a systematic comparison of different CNN architectures for automated classification of OSCC histopathological images within a unified experimental framework. Whole-slide images (WSIs) and patch-based analyses have been reported to capture morphological patterns beyond the visual perception of pathologists, thereby supporting diagnosis, grading, and even prognostic inference; however, barriers such as data standardization, validation design, and real-world distribution shifts (scanner/color/laboratory variation) remain major limitations [16–18]. In the context of oral cancer, recent studies reviewing deep learning applications across different data modalities (clinical images, histopathology, genomics, etc.) emphasize that insufficient data quality and the lack of external validation may inflate reported performance…
Editorial reading guide
- What data created and tested the model, and was evaluation independent?
- What is the cost of a false positive, false negative or out-of-distribution decision?
