Skip to main page content
U.S. flag

An official website of the United States government

Dot gov

The .gov means it’s official.
Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

Https

The site is secure.
The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2025 Aug 9;15(1):29218.
doi: 10.1038/s41598-025-11861-7.

Oral cancer detection via Vanilla CNN optimized by improved artificial protozoa optimizer

Yulong Chai  1 , Xiuqing Chai  2 , Lan Zhang  1 , Gang Ye  3 , Fatima Rashid Sheykhahmad  4   5
Affiliations

Oral cancer detection via Vanilla CNN optimized by improved artificial protozoa optimizer

Yulong Chai et al. Sci Rep. .

Abstract

In this study, we propose a new method for oral cancer detection using a modified Vanilla Convolutional Neural Network (CNN) architecture with incorporated batch normalization, dropout regularization, and a customized design structure for the convolutional block. An Improved Artificial Protozoa Optimizer (IAPO) metaheuristic algorithm is proposed to optimize the Vanilla CNN and the IAPO improves the original Artificial Protozoa Optimizer through a new search strategy and adaptive parameter tuning mechanism. Due to its effectiveness in search space exploration while avoiding local optimal points, the IAPO algorithm is chosen to optimize the convolutional neural network. In this study, a dataset of 1000 images of patients had published which will be preprocessed with contrast enhancement, noise reduction and data augmentation (like rotation, flipping and cropping) to generate the robust targeted model for oral cancer detection. The experimental results are evaluated against benchmark performance measures (accuracies, precision, recall, F1-score and area under the receiver operating characteristic (ROC) curve (AUC-ROC). We demonstrate through experimental results that the proposed IAPO optimized Vanilla CNN achieves a high accuracy of 92.5% which is superior than the previous state-of-the art models such as ResNet-101 (90.1%) and DenseNet-121 (89.5%). This proves to be a more trustworthy approach to oral cancer detectionbecause of the accuracy of the proposed method in comparison to denoting the supplementary results of the suggested method in contrast to other existing models.

Keywords: Cancer detection; Image classification; Improved artificial Protozoa optimizer; Metaheuristic optimization; Oral Cancer detection; Preprocessing; Vanilla convolutional neural networks.

PubMed Disclaimer

Conflict of interest statement

Declarations. Competing interests: The authors declare no competing interests.

Figures

Fig. 1
Fig. 1
Some samples of the OCI dataset.
Fig. 2
Fig. 2
Sample example of image resizing.
Fig. 3
Fig. 3
Sample example of image normalization.
Fig. 4
Fig. 4
Some sample examples of images augmentation.
Fig. 5
Fig. 5
The block diagram of the Vanilla CNN.
Algorithm 1
Algorithm 1
The improved APO algorithm.
Fig. 6
Fig. 6
The ablation analysis.
See this image and copyright information in PMC

Similar articles

See all similar articles

References

    1. Patel, S. & Kumar, D. Predictive identification of oral cancer using AI and machine learning. Oral Oncol. Rep.13, 100697 (2025).
    1. Mitbander, R. et al. Development and evaluation of an automated multimodal mobile detection of oral cancer (mDOC) imaging system to aid in risk-based management of oral mucosal lesions. Cancer Prev. Res. (2025). - PMC - PubMed
    1. Liu, H. & Ghadimi, N. Hybrid convolutional neural network and flexible dwarf mongoose optimization algorithm for strong kidney stone diagnosis. Biomed. Signal Process. Control. 91, 106024 (2024).
    1. Han, M. et al. Timely detection of skin cancer: an AI-based approach on the basis of the integration of echo state network and adapted seasons optimization algorithm. Biomed. Signal Process. Control. 94, 106324 (2024).
    1. Meena, G., Mohbey, K. K. & Kumar, S. Monkeypox recognition and prediction from visuals using deep transfer learning-based neural networks. Multimed. Tools Appl. 1–25 (2024).

LinkOut - more resources