Can artificial intelligence help clinicians in prostate cancer screening?
Principal Investigator
Name
Nicolas Martelin
Degrees
M.B.A., Ph.D.
Institution
Sacred Heart University
Position Title
Professor
Email
About this CDAS Project
Study
PLCO
(Learn more about this study)
Project ID
PLCO-345
Initial CDAS Request Approval
Feb 26, 2018
Title
Can artificial intelligence help clinicians in prostate cancer screening?
Summary
Prostate-specific antigen (PSA) has helped clinicians to detect prostate cancer at an early stage. The well-known biomarker has yet suffered from an going debate about its utility. PSA testing is indeed far from perfect since cut-off PSA levels ressemble a rule of thumb. When apply to the letter, those cut-off levels pave the way to overdiagnosis of indolent cancers, unnecessary treatments, and potential collateral health issues. This led the US Preventive Services Task Force to recommend against using PSA tests for routine prostate cancer screening.
To mitigate the drawbacks of PSA screening, serial PSA testing procedures have been introduced. PSA velocity and PSA doubling time measure the change of PSA level over time, making it easier for clinicians to detect a sudden rise in PSA levels. It is clearly a step towards a more personalized detection test. Yet those types of measurements are based on an over-simplification of a complex movement. PSA levels fluctuate up and down. Some causes of variation in PSA levels, outside of prostate cancer, have been identified such as climate, ejaculation, vigorous bike riding, etc. but some remain unexplained. PSA levels thus fluctuate randomly.
A non-random pattern can be detected by Artificial Intelligence (AI) tests long before a human clinician can even suspect it.
We propose to apply a non-randomness AI detection test to time series of PSA levels to determine whether it can help in detecting PSA cancers early without running the risk of overdiagnosis.
To mitigate the drawbacks of PSA screening, serial PSA testing procedures have been introduced. PSA velocity and PSA doubling time measure the change of PSA level over time, making it easier for clinicians to detect a sudden rise in PSA levels. It is clearly a step towards a more personalized detection test. Yet those types of measurements are based on an over-simplification of a complex movement. PSA levels fluctuate up and down. Some causes of variation in PSA levels, outside of prostate cancer, have been identified such as climate, ejaculation, vigorous bike riding, etc. but some remain unexplained. PSA levels thus fluctuate randomly.
A non-random pattern can be detected by Artificial Intelligence (AI) tests long before a human clinician can even suspect it.
We propose to apply a non-randomness AI detection test to time series of PSA levels to determine whether it can help in detecting PSA cancers early without running the risk of overdiagnosis.
Aims
Check wether AI test can find cancer or abnormality in patients with cancer (sensitivity)
Check wether Ai test can give negative results in patients with no cancer (specificity)
Check wether AI test can discriminate amongst fast and slow growing cancers
Collaborators
Brian DeWitt