Skip to Main Content

An official website of the United States government

Enhancing Prostate Cancer Detection from PSA Time-Series Using Recurrence Plots and Deep Learning.

Principal Investigator
Name
Alade Tokuta
Institution
North Carolina Central University
Position Title
Professor
Email
About this CDAS Project
Study
PLCO (Learn more about this study)
Project ID
PLCO-1945
Initial CDAS Request Approval
Jul 22, 2025
Title
Enhancing Prostate Cancer Detection from PSA Time-Series Using Recurrence Plots and Deep Learning.
Summary
This project investigates whether visualizing prostate-specific antigen (PSA) time-series data as recurrence plots and analyzing them with convolutional neural networks (CNNs) can improve the early detection of prostate cancer. Traditional PSA screening relies on fixed thresholds, often leading to missed diagnoses due to individual variability and benign fluctuations. By transforming PSA trends into images that capture temporal dynamics and training CNNs to classify these patterns, this research aims to uncover subtle, nonlinear indicators of cancer risk. The goal is to develop a more accurate and personalized tool for prostate cancer screening that improves on current methods in sensitivity and specificity.
Aims

This project aims to improve early prostate cancer detection by combining recurrence plot transformation with deep learning.

Aim 1 is to transform PSA time-series data into recurrence plots that visually capture the underlying dynamics, such as stability, abrupt changes, or chaotic patterns, that may signal malignancy.

Aim 2 is to train convolutional neural networks (CNNs) on these plots to classify PSA trends as either benign or cancerous.

Aim 3 is to compare this method’s performance against traditional machine learning models using raw or tabular PSA features.

Aim 4 is to interpret the learned features to better understand which PSA dynamics are most predictive, potentially revealing new biomarkers.

Collaborators

Misan Gift Esimaje - North Carolina Central University