Detection and Measurement of Lymph Nodes Using Artificial Intelligence; Unveiling the Impact of Artificial Intelligence in Medicine: Shaping Our Future

Introduction
The two primary documents in this portfolio are a technical report and a thesis research paper. The topics of the two are linked by the fact that they discuss issues relating to artificial intelligence in the field of medicine. The technical report is my lymph node detection tool using artificial intelligence and the STS paper is about the role that machine learning algorithms play in racial bias within the clinical field.

Technical Report: Detection and Measurement of Lymph Nodes Using Artificial Intelligence
The technical report in this portfolio is about the detection and measurement of lymph nodes using artificial intelligence algorithms. This paper starts by introducing the significance of cancer prevalence, specifically the rise of lymphoma disease, and some background around lymph nodes and current detection methods for lymphoma. It then discusses the proposed solution to relevant issues in detection and measurements, as well the intended aims for the project. After this introduction, there is an outline of results from the network models in breast cancer application and lung nodule scans. I go in depth to talk about the accuracy results of the models and comparing metrics across the trained machine learning models. Specific metrics that are talked about include validation and testing accuracy. Figure plots are made to display the accuracy of the sci-kit learning models, as well as the efficient network models. Also in this section, methods are outlined about how results were gathered and accuracy measurements were obtained. The report next touches on the safety and effectiveness of these models in radiological workflow, relating back to the aims that were mentioned in the introduction section.
Next, the paper goes on into the discussion section for analysis and interpretation of results. I begin to talk about the breast cancer machine learning application. The efficacy and potential of these efficient net models within radiological workflow are mentioned. There’s discussion about further improvements that could be made to the models in order to ensure reliability in clinical practice. Then, the paper goes on to talk about the lung nodule scans application with the 2.5D network model approach that was used with K fold cross validation testing. The paper mentioned the limitation of the dataset and its effect on the accuracy results.
Next, the paper delves into the radiomics model approach, interpreting the comparative accuracies to the 2.5D network model and other models. Similarly, I talk about the discrepancies in limited data sets as well as time constraints. The paper ends with talking about the future implications and studies that could be implemented with the machine learning models. It touches on how similar tasks should be worked on in the future and how this application serves as a great foundation for artificial intelligence algorithms in clinical practice.

STS Paper: Unveiling the Impact of Artificial Intelligence in Medicine: Shaping Our
Future
The paper begins with an introduction to the emergence of artificial intelligence and its current applications in many leading industries. Then, the paper touches on more specific applications for algorithms in clinical practice like diagnosis and detection. Next, I talk about how artificial intelligence can have profound effects on social factors like inequality and social injustice, affecting marginalized communities like African Americans. I then outline my research question that I attempt to answer using evidence from past historical biases and key texts. Next, the methods section of the paper outlines my approach for analysis using the Actor Network Theory, where I treat artificial intelligence as an actor and data as a player in the network of racial bias. The results section of my paper starts with general statistics and data that have shown racial bias within many fields for a long period of time. I dive into the Buolamwini and Gebru study on facial recognition systems to get the ball rolling on generalized bias within African American populations. Then, I talk more specifically about the data encrypted into these artificial intelligence tools by using evidence from my key text: “Medical Apartheid” by Harriet Washington. I use Washington’s qualitative analysis of historical biases to bolster my argument of biases in data that have been trained with these algorithms. I then back up this key text with quantitative analysis showing contemporary statistical data to show discriminatory bias in medical data and healthcare services. I move on to my next key text, “Breathing Race into the Machine” by Braun, L, where I talk further about the implications of historical bias with the spirometer. This evidence reinforces my argument of encoded bias into medical tools that underrepresented marginalized communities.
After presenting evidence of racial bias within medicine, I interpret my research and findings in the analysis and discussion sections. I mention the severity of implications that bias can have on African American communities today. I also talk about ways that this problem could be alleviated with methods of bias detection through algorithm auditing. Finally, the thesis discusses the issue of whose responsibility this problem is and some potential future oversights on the rising impact of artificial intelligence.

School of Engineering and Applied Science

Bachelor of Science in Biomedical Engineering

Technical Advisor: Xue Feng

STS Advisor: Joshua Earle

Technical Team Members: Nathan Patton, Daniel Song