Creative Commons source reading · Machine learning for performance
Texture-based image feature analysis for the classification of oral squamous cell carcinoma using machine learning approach
Salma Bahaa; Omneya Ramadan; Zeinab Elsayed Darwish; Eslam EL-Fiky; Mai M. Saleh
Originally published in BMC Oral Health · 2025-12-27
From the article
Selected source text. JournalOS removes markup and shortens excerpts where indicated. The full article provides the complete context.
Abstract
BackgroundArtificial intelligence is being developed as a vital tool to improve the accuracy and efficiency of oral cancer diagnosis. Oral squamous cell carcinoma is often detected at advanced stages, leading to poor patient outcomes, and traditional diagnostic methods can be subjective and physically demanding for pathologists. By comparing different Artificial intelligence models, including machine learning and deep learning, this study aims to create a more standardized and rapid diagnostic aid that can support pathologists and increase patient survival rates.MethodsA total of 115 cases (49 OSCC, 66 normal) were captured at 100x and 400x magnifications. Twelve Gray Level Co-occurrence Matrix (GLCM) features were extracted. Machine learning models, including Fine Tree, Medium Tree, Linear Discriminant, Linear Support Vector Machine,…
Clinical feature importance and ablation results
To mitigate the risk of patient-specific bias and confounding, we implemented two key procedural changes:3.Patient-Level Data Splitting: All cross-validation and final test sets were strictly separated at the patient level. This ensured that images from the same patient were never split across both the training and testing partitions. Table 8.4.Clinical-Only Model Robustness Check: To validate the discriminatory power of clinical features independent of image count per patient, we performed an additional test on the Clinical-Only model using a highly constrained dataset consisting of only one randomly selected image per patient. The maintenance of good performance in this restrictive setting is detailed in Additional File 8. The feature importance analysis, conducted using the Chi-Square (χ2) algorithm, confirmed Age and Anterior…
Discussion
Oral Squamous Cell Carcinoma (OSCC) represents approximately 90% of all oral neoplasms, yet its mortality rate has remained stubbornly high since the 1980 s [3–5]. While histopathological examination remains the gold standard for OSCC diagnosis and grading [8, 9], this manual process is often subjective, time-consuming, and prone to pathologist fatigue and inter-observer variability [10–12]. These limitations highlight a critical need for more precise, rapid, and standardized diagnostic methods to improve patient outcomes [6]. To address these challenges, we proposed the integration of AI to enhance the accuracy and efficiency of OSCC diagnosis [13]. Our study explored both traditional machine learning and deep learning approaches, leveraging AI’s ability to analyze medical images more objectively [27]. Specifically, we focused on…
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?
