Thesis Defence: Different Classifications Approaches Using Deep Learning and Hybrid Deep Neural Network to Quantify Agglutination Process Under Different Concentration
November 25 at 12:30 pm - 4:30 pm
Hridi Juberi, supervised by Dr. Ian Foulds, will defend their thesis titled “Different Classifications Approaches Using Deep Learning and Hybrid Deep Neural Network to Quantify Agglutination Process Under Different Concentration” in partial fulfillment of the requirements for the degree of Master of Applied Science in Electrical Engineering.
An abstract for Hridi Juberi’s thesis is included below.
Defences are open to all members of the campus community as well as the general public. Please email ian.foulds@ubc.ca to receive the Zoom link for this defence.
Abstract
Agglutination refers to the clustering of red blood cells, latex beads, or other microbeads which is crucial for biological processes and diagnostic assays, such as blood typing. However, quantifying agglutination accurately presents considerable difficulties, especially with different analyte concentrations. The droplets size, shape, aggregation, concentration, and interactions with other particles, makes the process more challenging. Moreover, traditional techniques for the quantification of agglutination frequently overlook the variations across testing conditions, including low and high analyte concentrations, thereby constraining precision and accuracy. In order to address these challenges and improve the accuracy of agglutination quantification, deep learning (DL) has been employed, owing to its capability to model non-linear relationships and discern complex patterns in large datasets. The study investigates the application of hybrid deep neural networks (HDNNs) to overcome the shortcomings of conventional methods of agglutination quantification by integrating the advantages from different Convolutional Neural Network (CNN) models. The study analyzed different experimental CNN models, such as, VGG16, EfficientNetB7, ResNet50, Xception, DenseNet121. In addition, a custom CNN model based on the architecture of ResNet and Xception was developed to evaluate its performance in comparison to the other CNN models analyzed previously. Lastly, a HDNN model was designed, integrating two CNN models, Xception and DenseNet121. According to the performance metrics, the HDNN model outperformed other CNN-based models and the custom CNN model providing a precise and reliable quantification across different concentration ranges.
