Abstract

In the domain of medical diagnostics, the precise classification of chest diseases has become very important, particularly during the occurrence of the COVID-19 pandemic. This project proposes an innovative approach to automate the classification process through the usage of deep learning technique, and more specifically making use of convolutional neural networks (CNNs) and transfer learning, with a prime focus on the RESNET50 architecture. The primary objective is to develop a model which is intelligent enough to detect various chest conditions, including COVID- 19, viral pneumonia, bacterial pneumonia, and normal cases. Utilizing the rigorously pre-trained knowledge encoded within RESNET50, trained on 11 million images containing 11,000 categories, leading to efficient transfer learning, a key strategy in addressing the challenges posed by the vanishing gradient problem encountered in deep learning models. Transfer learning enables the model to learn on its parameters and effectively make use of learned features from RESNET50, and establishing a robust model for the classification task while removing the risks associated with diminishing gradients. The transfer learning algorithm assigns appropriate weights and learning parameters to the input features, optimizing the model's accuracy and performance. The decision to implement RESNET50 over other transfer learning methods is supported by its proven efficiency and robustness in various computer vision tasks, majorly in medical imaging applications. RESNET50's architecture consists of 50 layers which allows for the extraction of hidden deep features from chest X-ray images which captures even the minute information which is necessary for accurate disease classification. Additionally, RESNET50's skip connections removes the vanishing gradient problem by allowing for effective information flow across layers and freezing the upper layers which holds the proper value for the weight and the biases, lead to enhanced model performance. The proposed methodology involves finetuning the pretrained RESNET50 model using a dataset comprising chest X-ray images which represents the target class. This fine- tuning process allows the model to adapt its parameters to the specific characteristics of chest disease classification, while retaining valuable features learned from the extensive dataset. During the implementation phase, a deep evaluation of the model's performance will be conducted, employing standard metrics such as accuracy, precision, recall, and F1-score. Furthermore, the model's performance will be assessed through cross-validation and testing on an independent validation dataset. Moreover, this project will explore the performance and efficiency of various optimizers, including Adam, Stochastic Gradient Descent (SGD), RMSprop, Adagrad, and Adamax, in fine-tuning the RESNET50 model. Through proper and in-depth comparison, it is expected that Adam optimizer will be more powerful in terms of performance, providing valuable insights into the impact of optimization algorithms on model training. The usage of this project for medical diagnostics are profound and it offers a reliable and effective tool for early detection and classification of chest diseases. By harnessing the power of deep learning and transfer learning techniques, healthcare professionals can largely benefit from fast and precise diagnoses, and therefore mitigating the risks on healthcare systems. In conclusion, the proposed approach represents an significant innovation in the domain of chest disease classification, and displaying the importance of deep learning and transfer learning methods to address complex medical challenges while side by side also optimizing model training for enhanced accuracy and efficiency.

Library of Congress Subject Headings

Chest--Diseases--Classification; Chest--Diseases--Diagnosis; Transfer learning (Machine learning)

Publication Date

12-10-2024

Document Type

Thesis

Student Type

Graduate

Degree Name

Professional Studies (MS)

Department, Program, or Center

Graduate Programs & Research

Advisor

Ehsan Rashedi

Campus

RIT Dubai

Plan Codes

PROFST-MS

Download

Share

COinS