[C10] Data Science and AI & Bioinformatics

[C10-02] Non-invasive Approach for the Characterization of Colorectal Cancer Using Health Check-up Data

*Kayla Gusti Haruni¹, Raiki Yoshimura¹, Takeru Matsuura¹, Shingo Iwami¹ (1. Nagoya University (Japan))

Keywords:

Colorectal cancer,Health check-up data,Machine learning

Colorectal cancer (CRC) is one of the most prevalent forms of cancer with high mortality. Currently, the most accurate diagnosis method is an endoscopy screening, which is expensive, invasive and detects the disease after the tumor’s formation. To address these limitations, this research proposes a non-invasive, machine learning-based approach for the prediction and characterization of CRC progression in patients using health check-up data. We utilized a large patient dataset from cohorts who underwent endoscopic examinations at Tokyo Women’s Medical University. The dataset includes non-invasive health check-up data such as clinical information and blood tests, along with corresponding endoscopy outcomes. Patients were labeled as either “healthy” or “polyp” based on the presence of polyps. Important features were identified using PowerSHAP, then a predictive classification model was developed using LightGBM which achieved high accuracy in distinguishing the characteristics between "healthy " and"polyp " cases. Beyond classification, the model also generated risk probabilities for all time points of each patient, enabling time series analysis that can visualize CRC progression. The result suggested that the risk probability time series exhibited a sigmoid-like pattern. By fitting a sigmoid function to each patient’s time series, parameters like slopes and inflection point can be extracted which reveal different rates of progression across the patients. Through this, we hope to isolate and analyze the time point representing early onset of the disease. These approaches may give valuable insights and novel perspectives into identifying patients who are at high-risk of developing CRC which can contribute towards developing non-invasive CRC diagnosis method and minimizing the number of individuals who require endoscopic screening. Furthermore, analysis of the disease progression may help to uncover early indicators of CRC, aiding in timely intervention and improved patient outcomes.