MSc Thesis Defense: A Comparative Study of Walrus and Grey Wolf Optimization Algorithms in Deep Transfer Learning for Colorectal Cancer Classification by Tauseef Ahmed

 

 

MSc Thesis Defense by:

Tauseef Ahmed

Date: March 03, 2026

Time:  11:00 AM

Location: Room # MH105 (Memorial Hall)

 

Abstract:

Colorectal cancer (CRC) remains one of the leading causes of cancer-related mortality worldwide, and early and accurate diagnosis is crucial for improving patient outcomes. In recent years, deep learning and transfer learning have shown promising potential in medical image analysis. This thesis investigates the enhancement of CRC classification by integrating metaheuristic optimization algorithms with pretrained convolutional neural networks (CNNs). Four widely used CNN architectures VGG16, VGG19, EfficientNetB0, ResNet-50 and DenseNet121 were employed as baseline models, trained on histopathological images of colorectal tissue. To further improve classification performance, two nature-inspired metaheuristic algorithms, the Walrus Optimization Algorithm (WOA) and the Grey Wolf Optimizer (GWO), were applied to optimize critical hyperparameters including learning rate, dropout rate, and dense layer units within defined bounds. Experimental results demonstrated that DenseNet121 consistently achieved the best performance across all experiments, with a baseline accuracy of 96.8%, which further improved to 97.6% with WOA optimization and 97.0% with GWO optimization. VGG16 and VGG19 also benefited from optimization, showing moderate improvements in accuracy and F1-scores, whereas EfficientNetB0 struggled to converge effectively in both baseline and optimized settings. Comparative analysis revealed that WOA slightly outperformed GWO in terms of stability and classification accuracy, particularly for DenseNet121. The findings highlight that combining deep transfer learning with metaheuristic hyperparameter optimization significantly enhances colorectal cancer classification performance.

Thesis Committee:

Internal Reader: Dr. Imran Ahmad           

External Reader: Dr. Faouzi Gherib          

Advisor: Dr. Boubakeur Boufama

Chair:    Dr. Jessica Chen