CST 499 Week 5

Parallel Progress to Unified Vision

This week marked a crucial phase in our Onco-Logic capstone project, as we shifted from independent development sprints to a focused integration effort. With core development largely complete across our three machine learning tracks, our primary goal was to weave these distinct components into a cohesive and impactful system.

Project Milestones Accomplished This Week

My personal contributions this week were heavily focused on consolidating our work and preparing for the final presentation. Here's what I accomplished:

• Developed the Streamlit demo app for the NLP project: I built an interactive front-end demonstration that showcases our ClinicalBERT-based pipeline. This app allows users to input pathology report text and see how our system extracts structured data, highlighting the practical application of our NLP work.

• Integrated prior model outputs into the main Onco-Logic system: This involved refining the preprocessing steps for the outputs from our breast cancer prognosis and genomic data projects. The goal was to ensure consistent data formats and seamless integration with the broader Onco-Logic framework.

• Resolved compute access issues: A significant hurdle we faced was inconsistent access to high-performance computing resources. I took the initiative to set up SSH connections to Google Colab, enabling more reliable access. Additionally, a teammate discovered that Colab now offers free access to Pro-level GPUs, which will be instrumental for running our more resource-intensive PyTorch and TensorFlow models.

• Contributed to the final report and presentation drafting: I began drafting sections of our technical report and presentation slides, specifically focusing on summarizing the methodologies and results of our NLP and cancer-subtype classification projects.

Plan for Next Week

Next week will be dedicated to solidifying our final deliverables and perfecting our presentation. Our key objectives include:

• Finalizing the technical report: We'll be completing the writing and formatting of our comprehensive technical report, ensuring it accurately reflects our methodologies, findings, and the overall impact of Onco-Logic.

• Developing visuals for the presentation: I'll be working on creating clear, impactful visuals and slides for our final presentation, focusing on conveying complex information concisely and engagingly.

• Building out demo apps for survival prediction models: We aim to create interactive Streamlit demos for our breast cancer prognosis models, allowing users to explore survival predictions based on clinical data.

• Conducting internal reviews: We'll perform thorough internal reviews of all integrated components and demo applications to ensure smooth functionality and a polished user experience. This will help us identify and address any remaining bugs or inconsistencies.

Challenges Faced

While we've made significant progress, we've encountered a couple of challenges:

• Compute access: This was a substantial roadblock for much of our development, particularly for training and running resource-heavy models. While we've largely mitigated this through the strategic use of Google Colab and its newly available Pro-level GPUs, it required extra effort to establish stable environments.

• Aligning outputs from different modules: Because each team member initially worked independently on distinct projects, integrating their respective outputs into a unified system has required careful coordination and some re-engineering of preprocessing and formatting steps to ensure consistency.

Looking Ahead

We're now on the cusp of completing a truly comprehensive system that spans structured clinical data, high-dimensional genomic inputs, and free-text analysis from pathology reports. The initial decision to work independently, though counterintuitive at first, has undeniably paid dividends, allowing us to explore a wider range of solutions and refine individual expertise. Now, our collective focus is on refining, integrating, and presenting a complete, functional AI pipeline for oncology that demonstrates the tangible benefits of machine learning in this critical domain.