Comparing the Predictions of Convolutional Neural Networks, Random Forest Transfer Learning, and Support Vector Machines in the Image Processing of Neoplasms
School Name
Spring Valley High School
Grade Level
11th Grade
Presentation Topic
Computer Science
Presentation Type
Non-Mentored
Abstract
This study investigated the efficacy of Convolutional Neural Networks (CNNs), Support Vector Machines (SVMs), and Random Forest Transfer Learning (RFTL) in the image processing of lung neoplasms for enhanced diagnostic accuracy. Leveraging the Lung Image Database Consortium Image Database Resource Initiative (LIDC-IDRI), chest CT images were analyzed to compare the three models' performance. It was hypothesized that the most accurate predictions would result from the RFTL model because it would have the most efficient data processing. Results indicated that all three models demonstrated high accuracy, with CNNs excelling in precision, SVMs in recall, and RFTL showing a balanced performance. Precision-recall curves highlighted trade-offs between models, emphasizing the importance of considering diagnostic priorities. Comparisons with existing literature and models revealed similarities and differences, underlining the adaptability of these machine learning approaches to different medical imaging tasks. Despite notable achievements, the study acknowledged limitations, such as hardware constraints and environmental variations during training. Recommendations included the exploration of ensemble methods, hyperparameter tuning, and data augmentation for improved model performance. Future research avenues involved testing alternative CNN models, diverse machine learning algorithms, and the inclusion of clinical data for a more comprehensive analysis. In conclusion, this research contributed insights into the comparative strengths and trade offs of CNNs, SVMs, and RFTL in lung cancer image diagnosis. The findings underscored the models' potential clinical relevance and provided a foundation for refining techniques in medical image processing, paving the way for improved early detection and patient outcomes in lung cancer diagnosis.
Recommended Citation
Wallam, Ali, "Comparing the Predictions of Convolutional Neural Networks, Random Forest Transfer Learning, and Support Vector Machines in the Image Processing of Neoplasms" (2024). South Carolina Junior Academy of Science. 474.
https://scholarexchange.furman.edu/scjas/2024/all/474
Location
RITA 367
Start Date
3-23-2024 10:45 AM
Presentation Format
Oral and Written
Group Project
No
Comparing the Predictions of Convolutional Neural Networks, Random Forest Transfer Learning, and Support Vector Machines in the Image Processing of Neoplasms
RITA 367
This study investigated the efficacy of Convolutional Neural Networks (CNNs), Support Vector Machines (SVMs), and Random Forest Transfer Learning (RFTL) in the image processing of lung neoplasms for enhanced diagnostic accuracy. Leveraging the Lung Image Database Consortium Image Database Resource Initiative (LIDC-IDRI), chest CT images were analyzed to compare the three models' performance. It was hypothesized that the most accurate predictions would result from the RFTL model because it would have the most efficient data processing. Results indicated that all three models demonstrated high accuracy, with CNNs excelling in precision, SVMs in recall, and RFTL showing a balanced performance. Precision-recall curves highlighted trade-offs between models, emphasizing the importance of considering diagnostic priorities. Comparisons with existing literature and models revealed similarities and differences, underlining the adaptability of these machine learning approaches to different medical imaging tasks. Despite notable achievements, the study acknowledged limitations, such as hardware constraints and environmental variations during training. Recommendations included the exploration of ensemble methods, hyperparameter tuning, and data augmentation for improved model performance. Future research avenues involved testing alternative CNN models, diverse machine learning algorithms, and the inclusion of clinical data for a more comprehensive analysis. In conclusion, this research contributed insights into the comparative strengths and trade offs of CNNs, SVMs, and RFTL in lung cancer image diagnosis. The findings underscored the models' potential clinical relevance and provided a foundation for refining techniques in medical image processing, paving the way for improved early detection and patient outcomes in lung cancer diagnosis.
