Related Experiment Video
Updated: Jun 11, 2025
07:30
Learning Modern Laryngeal Surgery in a Dissection Laboratory
Published on: March 18, 2020
8.0K
Exploring the Impact of Model Complexity on Laryngeal Cancer Detection.
Divya Rao1, Rohit Singh2, Prakashini Koteshwara3
1Department of Information and Communication Technology, Manipal Institute of Technology, Manipal Academy of Higher Education, Manipal, 576104 India.
Summary
This study compared machine learning models for laryngeal cancer detection. Deep learning models like ResNet-50 showed superior accuracy in identifying tumors on CT scans compared to simpler methods.
Area of Science:
- Medical imaging analysis
- Artificial intelligence in oncology
- Computational pathology
Background:
- Laryngeal cancer is a significant subset of head and neck cancers, making early detection crucial for treatment success.
- Machine learning (ML) offers potential for medical diagnostics, but the influence of ML model complexity on laryngeal cancer detection accuracy requires further investigation.
- Computed tomography (CT) imaging is a key modality for laryngeal cancer diagnosis.
Purpose of the Study:
- To investigate the relationship between machine learning model complexity and diagnostic efficacy in laryngeal cancer detection using CT images.
- To compare the performance of Logistic Regression, a shallow neural network, and a deep convolutional neural network (ResNet-50) for this task.
- To provide insights into selecting appropriate ML models for laryngeal cancer diagnosis based on complexity and performance.
Main Methods:
- Evaluation of three distinct machine learning models: Logistic Regression, a 4-layer Neural Network (NN), and a 50-layer Convolutional Neural Network (ResNet-50).
- Application of these models to identify laryngeal cancer from computed tomography (CT) images.
- Quantitative assessment of diagnostic accuracy for each model.
Main Results:
- Logistic Regression achieved an accuracy of 82.5%.
- The 4-Layer NN demonstrated improved accuracy at 87.2%.
- ResNet-50, a deep learning model, attained the highest accuracy of 92.6%, attributed to its advanced feature extraction capabilities on CT images.
Conclusions:
- Model selection for laryngeal cancer detection involves trade-offs between interpretability, complexity, and accuracy.
- While Logistic Regression is interpretable, it may not capture intricate patterns effectively.
- Deep learning models like ResNet-50 offer superior performance for complex image analysis in laryngeal cancer detection but require substantial computational resources.
