Combining a Risk Factor Score Designed From Electronic Health Records With a Digital Cytology Image Scoring System to...

Area of Science:

• Oncology
• Medical Informatics
• Machine Learning in Healthcare

Background:

• Bladder cancer mortality necessitates improved early detection methods.
• Current diagnostic tools lack non-invasiveness, affordability, and high performance.
• Cytology image classification models show promise but require further enhancement.

Purpose of the Study:

• To evaluate combining clinical risk factor data with a digital cytology image model (VisioCyt) for bladder cancer detection.
• To develop a robust bladder cancer scoring system using integrated clinical and image-based data.

Main Methods:

• Developed a predictive model using clinical data and machine learning algorithms (logistic regression, random forest, SVM).
• Extracted risk factors from a hospital's clinical data warehouse.
• Investigated strategies to combine the clinical risk score with the VisioCyt image-based model score.

Main Results:

• A risk factor-based model was designed using data from 5422 patients.
• The combined model achieved an Area Under the Curve (AUC) of 0.82 (training) and 0.83 (test) on a 620-patient dataset.
• The combined approach demonstrated higher cancer risk association than VisioCyt alone, especially for low-grade bladder cancer.

Conclusions:

• Combining clinical and biological information significantly enhances bladder cancer detection, particularly for low-grade tumors.
• Further refinement of automatic clinical feature extraction is needed for model robustness.
• The integrated approach shows potential benefit for patient outcomes.