The use of an AI software tool for prostate volume measurement on MRI
The use of an AI software tool for prostate volume measurement on MRI
P Ward, M Sakthithasan, E Lambert, S Sangma, J Kasznia-Brown, PR Burn
To determine the effectiveness of an AI software tool for measurement of prostate volume on MRI. Prostate cancer is the most common cancer in UK men, with around 52,300 new diagnoses/year, representing 27% of all new cancer cases in males (2016/18) [1]. At our institution1 referrals have increased by 50% between 2019 and 2022 and <1% of our patients met the UK NHS’s 28 day diagnostic pathway target (MRI and biopsy) for the past year [2]. PiTM (Prostate Intelligence, Lucida Medical) is an AI-based software product for prostate cancer diagnosis on MRI. We have been evaluating Pi at our institution as a possible aid to improving workflow within our diagnostic pathway.
Participants Consecutive patients were included who had previously been investigated for suspected prostate cancer with MRI. Data Collection and Analysis All patients had been scanned at 3T (Siemens Vida 3T (60/200 XT gradients). Part One For each patient, the radiologist’s calculated prostate volume given in the clinical report was recorded. Radiologists’ volumes were measured using the ellipsoid formula and following PI-RADS v2.1 recommendations3. Prostate volumes were then measured on MRI using Pi (version 2.2, Lucida Medical). Radiologists’ and Pi prostate volumes were compared (Pearson correlation coefficient). The absolute percentage difference* (APD) for each patient, and the mean APD for the group, were calculated. Bland-Altman analysis was also performed, plotting the percentage difference between Pi and the radiologist’s volume, against mean prostate volume*, for each patient. Part Two For the second part of the study, 30 patients from the above cohort were randomly selected and 4 radiologists asked to independently re-measure the prostate volume. The mean APD of the 30 prostate volumes between 6 pairs of radiologists, and between Pi and each of the 4 radiologists, was calculated. *Definitions: (for two measurements of prostate volume in a patient, x and y). Absolute percentage difference (APD) = 100 * |x-y|/(x+y)/2 Mean prostate volume = (x+y)/2
Comparison between Pi volume and radiologist measured prostate volume (628 patients)
Bland Altman plot of percentage difference between Pi and radiologist’s volume against mean prostate volume (628 patients)
Table showing for 30 patients, mean APD (and range) between pairs of radiologists (A to D) and between Pi and each radiologist.
Part One: 628 patients. Mean prostate volume of group (range): radiologist 61ml (16 - 330ml), Pi 59ml (11- 325ml). Pi and radiologist volume measurements were strongly positively correlated, r(626)= .958, p<0.001 (see first graph). Mean APD (range) between radiologist and Pi volume was 15.5% (0-92%). Bland-Altman plot: bias -6.1%, 95% CIs -44 to +31% (see second graph). Part Two: See table, showing for 30 patients, mean APD (and range) between pairs of radiologists (A to D) and between Pi and each radiologist.
There is strong correlation between Pi and radiologist’s prostate volumes taken from clinical practice, with a mean absolute percentage difference of 16%. In our smaller cohort of 30 study patients, agreement between Pi and radiologists is somewhat better (mean absolute percentage difference 11-13%), which might reflect more careful measurements by radiologists, and is comparable to inter-radiologist agreement. Pi volume measurements are considered suitable for routine use in clinical practice. References: 1. https://www.cancerresearchuk.org/health-professional/cancer-statistics/statistics-by-cancer-type/prostate-cancer 2. Southwest UK Prostate Dashboard – contact Paul.Burn@nhs.net for details 3. https://www.acr.org/-/media/ACR/Files/RADS/Pi-RADS/PIRADS-v2-1.pdf Acknowledgements: Kate Hadley and Stacey Thorne for many hours of admin support
