Automatic Detection in MRI of Prostate Cancer: DAICAP
- Conditions
- Detection and Characterization of Prostate Cancer Based on Artificial Intelligence
- Registration Number
- NCT05513820
- Lead Sponsor
- Assistance Publique - Hôpitaux de Paris
- Brief Summary
Prostate cancer is the most common cancer in France and the 3rd most common cancer death in humans. The introduction of pre-biopsy MRI has considerably improved the quality of prostate cancer (PCa) diagnosis by increasing the detection of clinically significant PCa , and by reducing the number of unnecessary biopsies.However the diagnostic performance of Prostate MRI is highly dependent on reader experience that limits the population based delivery of high quality multiparametricMRI (mpMRI) driven PCa diagnosis. The main objective of this study is the development and the test of diagnostic accuracy of an AI algorithm for the detection of cancerous prostatic lesions from mpMRI images.
The secondary objective is the development and the test of diagnostic accuracy of an AI algorithm to predict tumor aggressiveness from mpMRI images.
- Detailed Description
This is a study combining :
1. Firstly a sub-study with a multicentric retrospective sample of 700 patients from the databases of AP-HP, CHU de Lyon and CHU de Lille for training and validation of algorithms. The historical depth may be up to 96 months (8 years)
2. A second sub-study with a multicentric prospective sample of 550 patients (test-set) associating AP-HP (CHU Pitié, Tenon, Bicêtre, Necker), CHU Lille, CHU Lyon, CHU Bordeaux and CHU Strasbourg to test the performance of algorithms Data will be collected retrospectively (training phase - validation of the algorithm) and prospectively (testing phase of the algorithm) from the medical records of each of the centres for patients corresponding to the inclusion and exclusion criteria mentioned above.
Methodology :
1. Retrospective phase mpMRI images chained to histological (prostate biopsy data), biological (PSA) and demographic (age) data will be used for supervised learning during the training and validation phases. Thus, the aggressiveness scores will rely on a matching between mpMRI images and the results of targeted biopsies in addition to standard biopsies
2. Prospective phase For the performance measurement, a test set of 550 prospectively collected images will be used, of which 150 will be from the same centers, and 400 from 3 other clinical centers (CHU Strasbourg, APHP Bicêtre and Necker-HEGP and CHU Bordeaux).
The algorithms developed in the retrospective phase will be applied by Inria to the prospective data, without knowledge of the PI-RADS score or the aggressiveness. The performance of each algorithm will then be evaluated, under the responsibility of an independent unit,by its sensitivity and specificity with their IC95%. The main analysis will be conducted by patient (presence of at least one lesion with a PI-RADS score ≥3; presence of at least one lesion considered aggressive (defined by the presence of a histological Gleason score grade 4 up to 6 months after the mpMRI). Secondary analyses will be conducted by lesion and by prostate lobe.
Recruitment & Eligibility
- Status
- RECRUITING
- Sex
- Male
- Target Recruitment
- 1250
- Patients with clinical suspicion of prostate cancer (increased PSA and/or abnormality on digital rectal examination) who underwent a diagnostic workup including mpMRI and prostate biopsies according to national recommendations: in case of normal mpMRI (PI-RADS < 3) 12 systematic samples; in case of pathological mpMRI (PI-RADS ≥3) 12 systematic samples associated with targeted samples (n= 2 to 4) by cognitive fusion, or image fusion software.
- Patients with histologically proven prostate cancer and/or treatment for prostate cancer prior to the diagnostic workup
Prospective substudy
Inclusion Criteria:
- Patients with clinical suspicion of prostate cancer (increased PSA level and/or abnormality on digital rectal examination) who should receive a diagnostic workup including mpMRI and prostate biopsies according to national recommendations: in case of normal mpMRI (PI-RADS < 3) 12 systematic samples; in case of pathological mpMRI (PI-RADS ≥3) 12 systematic samples associated with targeted samples (n= 2 to 4) by cognitive fusion, or image fusion software.
Exclusion Criteria:
- Patients with already histologically proven cancer, patients who have received treatment for prostate cancer, patients who cannot benefit from prostate biopsies, or patients with a contraindication to performing mpMRI.
Study & Design
- Study Type
- OBSERVATIONAL
- Study Design
- Not specified
- Primary Outcome Measures
Name Time Method Performance (sensitivity and specificity) of the algorithm to predict the standardized radiological PI-RADS Inclusion The primary endpoint will be the performance (sensitivity and specificity) of the algorithm to predict the standardized radiological PI-RADS score for each patient: presence of at least one lesion considered significant (internationally standardized score between 1 and 5 and with a threshold of positivity at 3 or more).
- Secondary Outcome Measures
Name Time Method Performance (sensitivity and specificity) of the algorithm in predicting tumor aggressiveness Inclusion The secondary endpoint will be the performance (sensitivity and specificity) of the algorithm in predicting tumor aggressiveness.Gold standard is defined, for each patient, on the histological analysis of the first series of biopsy samples taken in the course of care,up to 6 months from the mpMRI, as the presence of at least one lesion with grade 4 cells according to the characterization by the international histoprognostic score ISUP.Patients without biopsy in the 6 months will be considered as having non aggressive tumor.
Trial Locations
- Locations (1)
La Pitié Salpétrière Hospital
🇫🇷Paris, France