Computer Science Department Thesis Defense - Aya Farrag

Please join the Computer Science Department for the upcoming thesis defense:

Presenter: Aya Farrag

Thesis title: Towards a Trustworthy End-to-End AI-based Clinical Decision Support System for Disease Diagnosis and Prognosis: Breast Cancer Use-Case


Abstract: Artificial Intelligence (AI) research has emerged as a powerful tool for health-related applications. With the increasing shortage of radiologists and oncologists around the world, developing an end-to-end AI-based Clinical Decision Support (CDS) system for fatal disease diagnosis and survivability prediction can have a significant impact on healthcare professionals as well as patients. Such a system uses machine learning algorithms to analyze medical images and clinical data to detect cancer, estimate its survivability and aid in treatment planning. We can break the CDS system down into two main components: the Computer Aided Diagnosis (CAD) subsystem and the Computer-Aided Prognosis subsystem (CAP). The lack of trustworthiness of these systems is still considered a challenge that needs to be addressed in order to increase their adoption and usefulness in real-world applications.

In this thesis, we propose new methods and frameworks to address existing challenges and research gaps in both components of the system to pave the way toward its usage in clinical practice. We select the breast cancer use case to perform our analysis as it is the most common cancer amongst women and the second leading cause of cancer death.

In cancer CAD systems, the first and most important step is to analyze medical images to identify potential tumors in a specific organ. In dense prediction problems like mass segmentation, preserving the input image resolution plays a crucial role in achieving good performance. However, this resolution is often reduced in current Convolution Neural Networks (CNN) that are commonly repurposed for this task. In Chapter 3, we propose a double-dilated convolution module in order to preserve spatial resolution while having a large receptive field. The proposed module is applied to the tumor segmentation task in breast cancer mammograms as a proof-of-concept. To address the pixel-level class imbalance problem in mammogram screenings, different loss functions (i.e., binary cross-entropy, weighted cross-entropy, dice loss, and Tversky loss) are evaluated. Experimental analysis is performed to compare the performance of lesion segmentation networks on mammogram screenings before and after plugging the proposed module into one state-of-the-art deep convolutional neural network. The obtained results show the effectiveness of the proposed module in increasing the similarity score and decreasing the miss-detection rate.

Following the cancer diagnosis step, in Chapter 4, we propose a new framework for cancer survival prediction in CAP systems to precisely predict the estimated survival months of patients in order to facilitate treatment planning. We combine two main strategies in solving the cancer survivability prediction problem using Machine Learning techniques. In the first strategy, we model the survivability prediction task as a two-step problem, namely a classification problem to predict whether or not a patient survives for five years, and a regression problem to forecast the number of remaining months for those who are predicted to not survive for five years. The second strategy is to develop stage-specific models, where each model is trained on instances belonging to a certain cancer stage in order to precisely predict survivability of patients from the same stage. We investigate the impact of adopting these strategies along with applying different balancing techniques over the model performance using breast cancer clinical data. The obtained results demonstrate that the proposed methods prove effective in both survivability classification and regression.

To incorporate the role of prognosis in determining the most suitable treatment plans for a cancer patient, in Chapter 5, we propose a novel prognostic-based framework for treatment planning. We employ the prediction models developed for stage-specific survival estimation to determine the best possible treatment plans in terms of prognostic outcomes. The system generates an ordered list of all possible combinations of treatments associated with their predicted survival outcomes to offer more comprehensive and intuitive treatment recommendations that can aid medical professionals in making more informed decisions about the most appropriate course of treatment for their patients. By integrating survival prediction models into treatment planning, healthcare providers can offer better patient care and help patients and their families make more informed decisions about their treatment options.

Experiments conducted in different chapters of this thesis demonstrate that the proposed AI-enabled techniques may potentially be utilized in future Clinical Decision Support Systems to help clinicians make patient-specific assessments and treatment decision



Committee Members:
Dr. Zubair Fadlullah (supervisor, committee chair), Dr. Muhammad Asaduzzaman, Dr. Al-Sakib Khan Pathan (United International University)

Please contact grad.compsci@lakeheadu.ca for the Zoom link.
Everyone is welcome.