Ph. D. Project
Title:
Validation of a prostate tumor risk calculator, with a new personalized PSA analysis model and lifestyle based
on artificial intelligence
Dates:
2024/10/01 - 2027/09/30
Supervisor(s):
Other supervisor(s):
Pr FRADET Vincent (vincent.fradet@fmed.ulaval.ca)
, Nicolas Martelin (nicolas.martelin@prostperia.com)
Description:
Early detection of prostate cancer remains a major public health issue worldwide. With this in mind, there
has been growing interest in using artificial intelligence (AI) to assess prostate tumor risk based on PSA
analysis. This research work focuses on the validation of an AI-based risk calculator, offering a novel
perspective for improving screening accuracy.
I. Background
Prostate cancer remains one of the most common cancers in men, and its early detection is essential to
improve the chances of successful treatment. PSA, a commonly used marker, can be exploited in a more
advanced way thanks to AI.
The literature has already shown the importance of identifying prostate cancer patients before the cancer
becomes symptomatic. Large randomized clinical trials (Göteborg, ERSPC, PLCO) have demonstrated the
benefits and limitations of PSA screening on cancer-specific survival.
Prostate cancer screening is mainly based on PSA testing. However, due to its lack of specificity, this test
also leads to over-treatment of slowly progressing cancers.
The integration of AI in healthcare offers opportunities for personalized diagnosis and risk prediction. An
AI-based risk calculator for prostate cancer could help reduce false alarms and identify at-risk cases more
accurately.
II. Methodology
1. PROSTia test
To improve the specificity of screening, several tests have recently been developed, such as the PHI test or
the 4Kscore test 4. The most recent of these tests is PROSTia. This is an "in silico" test that personalizes the
interpretation of the PSA serological value thanks to the probabilistic analysis of several Machine-Learning
algorithms. A series of classification runs using the Gradient Boosting technique optimize the model's
hyper-parameters, resulting in the calculation of a final score: the PROSTia test result.
PROSTia is a predictive test for prostate cancer risk, based on PSA analysis, changes over time, digital rectal
examination (DRE) and over 50 personal parameters. Among these parameters, PROSTia takes into account
the patient's family, medical and medication history.
PROSTia has been validated on a retrospective cohort of 12,000 patients (PLCO), the results of which are
currently being published.
An abstract presenting the initial results of the PROSTia algorithm will be presented as a poster at the EAU
in Paris in April 2024.
An ongoing study at the Nancy CHRU shows that the test could also be used as a biopsy decision aid,
leading to a potential 60% reduction in unnecessary biopsies.
3. Data collection in Canada
In recent years, the literature has begun to show the importance of lifestyle and diet in the prevention or
occurrence of prostate cancer.
Prof. Fradet's team at Laval University is developing work on this subject. We hypothesize that the PROSTia
test can be improved by adding parameters on lifestyle, perceived quality of life and diet to its predictive
functions.
Université Laval (Quebec) is leading a multicenter study on prostate cancer diagnosis, called BioCaPPE .
The study uses a prospective multi-institutional pan-Quebec design to evaluate biomarkers of prostate
cancer risk in relation to lifestyle habits. It includes over 2,050 participants recruited from 5 sites across the
province of Quebec. Data collection at study entry uses several validated questionnaires to measure
potentially modifiable lifestyle habits, including physical activity and nutrition.
4. Model training
The PROSTia test will be trained with machine learning algorithms using the data collected. All subjects
from the BioCappe cohort will be analyzed with the PROSTia test. The aim will be to validate its
effectiveness in detecting prostate cancer in this Canadian population. Using the other data collected,
particularly those relating to lifestyle habits, we will also try to optimize the model's performance.
III. Validation of the Risk calculator
1. Validation parameters
Accuracy, sensitivity and specificity will be assessed to measure the effectiveness of the risk calculator.
Receiver Operating Characteristic (ROC) curves will be generated to assess overall model performance.
2. Comparison with Traditional Methods
The results of the AI-based risk calculator will be compared with traditional prostate cancer screening
methods to demonstrate its advantage in terms of accuracy and predictive ability.
V. Objectives of my research :
Under the supervision of Professor Fradet's team (Université Laval), my objectives are to study PROSTia in
these two configurations:
- Use the test on the BioCaPPE cohort to evaluate the test's performance on a Canadian population.
- PROSTia currently includes some lifestyle-related data, but it may be possible to further improve its
predictive accuracy. To this end, we will be working on a new version of the algorithm that will include a
greater wealth of lifestyle data and related biomarkers.
VI. Conclusion
In conclusion, the validation of a prostate tumor risk calculator, based on PSA analysis using artificial
intelligence, offers a promising prospect for improving screening practices. The expected results could
make a significant contribution to the advancement of prostate cancer diagnosis methods, paving the way
for a more precise and personalized approach to men's health.
This work will also create an international link with Professor Vincent FRADET's team.
has been growing interest in using artificial intelligence (AI) to assess prostate tumor risk based on PSA
analysis. This research work focuses on the validation of an AI-based risk calculator, offering a novel
perspective for improving screening accuracy.
I. Background
Prostate cancer remains one of the most common cancers in men, and its early detection is essential to
improve the chances of successful treatment. PSA, a commonly used marker, can be exploited in a more
advanced way thanks to AI.
The literature has already shown the importance of identifying prostate cancer patients before the cancer
becomes symptomatic. Large randomized clinical trials (Göteborg, ERSPC, PLCO) have demonstrated the
benefits and limitations of PSA screening on cancer-specific survival.
Prostate cancer screening is mainly based on PSA testing. However, due to its lack of specificity, this test
also leads to over-treatment of slowly progressing cancers.
The integration of AI in healthcare offers opportunities for personalized diagnosis and risk prediction. An
AI-based risk calculator for prostate cancer could help reduce false alarms and identify at-risk cases more
accurately.
II. Methodology
1. PROSTia test
To improve the specificity of screening, several tests have recently been developed, such as the PHI test or
the 4Kscore test 4. The most recent of these tests is PROSTia. This is an "in silico" test that personalizes the
interpretation of the PSA serological value thanks to the probabilistic analysis of several Machine-Learning
algorithms. A series of classification runs using the Gradient Boosting technique optimize the model's
hyper-parameters, resulting in the calculation of a final score: the PROSTia test result.
PROSTia is a predictive test for prostate cancer risk, based on PSA analysis, changes over time, digital rectal
examination (DRE) and over 50 personal parameters. Among these parameters, PROSTia takes into account
the patient's family, medical and medication history.
PROSTia has been validated on a retrospective cohort of 12,000 patients (PLCO), the results of which are
currently being published.
An abstract presenting the initial results of the PROSTia algorithm will be presented as a poster at the EAU
in Paris in April 2024.
An ongoing study at the Nancy CHRU shows that the test could also be used as a biopsy decision aid,
leading to a potential 60% reduction in unnecessary biopsies.
3. Data collection in Canada
In recent years, the literature has begun to show the importance of lifestyle and diet in the prevention or
occurrence of prostate cancer.
Prof. Fradet's team at Laval University is developing work on this subject. We hypothesize that the PROSTia
test can be improved by adding parameters on lifestyle, perceived quality of life and diet to its predictive
functions.
Université Laval (Quebec) is leading a multicenter study on prostate cancer diagnosis, called BioCaPPE .
The study uses a prospective multi-institutional pan-Quebec design to evaluate biomarkers of prostate
cancer risk in relation to lifestyle habits. It includes over 2,050 participants recruited from 5 sites across the
province of Quebec. Data collection at study entry uses several validated questionnaires to measure
potentially modifiable lifestyle habits, including physical activity and nutrition.
4. Model training
The PROSTia test will be trained with machine learning algorithms using the data collected. All subjects
from the BioCappe cohort will be analyzed with the PROSTia test. The aim will be to validate its
effectiveness in detecting prostate cancer in this Canadian population. Using the other data collected,
particularly those relating to lifestyle habits, we will also try to optimize the model's performance.
III. Validation of the Risk calculator
1. Validation parameters
Accuracy, sensitivity and specificity will be assessed to measure the effectiveness of the risk calculator.
Receiver Operating Characteristic (ROC) curves will be generated to assess overall model performance.
2. Comparison with Traditional Methods
The results of the AI-based risk calculator will be compared with traditional prostate cancer screening
methods to demonstrate its advantage in terms of accuracy and predictive ability.
V. Objectives of my research :
Under the supervision of Professor Fradet's team (Université Laval), my objectives are to study PROSTia in
these two configurations:
- Use the test on the BioCaPPE cohort to evaluate the test's performance on a Canadian population.
- PROSTia currently includes some lifestyle-related data, but it may be possible to further improve its
predictive accuracy. To this end, we will be working on a new version of the algorithm that will include a
greater wealth of lifestyle data and related biomarkers.
VI. Conclusion
In conclusion, the validation of a prostate tumor risk calculator, based on PSA analysis using artificial
intelligence, offers a promising prospect for improving screening practices. The expected results could
make a significant contribution to the advancement of prostate cancer diagnosis methods, paving the way
for a more precise and personalized approach to men's health.
This work will also create an international link with Professor Vincent FRADET's team.
Keywords:
risk calculator; screening; prostate cancer; artificial intelligence; machine learning
Conditions:
The university thesis will last 3 years. It will start at the beginning of November 2024 with a period of
mobility abroad.
1st year :
mobility with Pr Vincent Fradet at CHRU Québec, Université de Laval , Canada
Funding applied for: Association des chefs de service CHRU Nancy research grant, AFU grant, CIFRE grant.
2nd and 3rd year :
ED BioSE Biologie, Santé, Environnement CRAN UMR 7039, CNRS, Liquid biopsies and therapeutic
optimization, Pr Stéphanie GRANDEMANGE
Funding: PHU status University of Lorraine and CHRU NANCY
mobility abroad.
1st year :
mobility with Pr Vincent Fradet at CHRU Québec, Université de Laval , Canada
Funding applied for: Association des chefs de service CHRU Nancy research grant, AFU grant, CIFRE grant.
2nd and 3rd year :
ED BioSE Biologie, Santé, Environnement CRAN UMR 7039, CNRS, Liquid biopsies and therapeutic
optimization, Pr Stéphanie GRANDEMANGE
Funding: PHU status University of Lorraine and CHRU NANCY
Department(s):
| Biology, Signals and Systems in Cancer and Neuroscience |
Funds:
1st year: scholarship , 2nd and 3rd year: PHU
