AUTHOR=Baldeon-Calisto Maria , Wei Zhouping , Abudalou Shatha , Yilmaz Yasin , Gage Kenneth , Pow-Sang Julio , Balagurunathan Yoganand 

TITLE=A multi-object deep neural network architecture to detect prostate anatomy in T2-weighted MRI: Performance evaluation

JOURNAL=Frontiers in Nuclear Medicine

VOLUME=Volume 2 - 2022

YEAR=2023

URL=https://www.frontiersin.org/journals/nuclear-medicine/articles/10.3389/fnume.2022.1083245

DOI=10.3389/fnume.2022.1083245

ISSN=2673-8880

ABSTRACT=We present a 2D-3D convolutional neural network (CNN) ensemble automatically constructed to segment the whole prostate gland along with the peripheral zone (PZ) (PPZ-SegNet) using T2-weighted sequence of MRI. The study used four different data sets all obtained from The Cancer Imaging Archive (TCIA). The prostate gland and the zonal anatomy (PZ) were manually delineated by consensus read through a clinical reader, using the axial T2W for three data sets. In this work, the PPZ-SegNet was trained using 150 cases of ProstateX data with prostate and PZ annotations. The model evaluation was performed on an independent cohort of 283 T2W MRI prostate cases, obtained from a diverse cohort. The segmentation performance was evaluated by computing the dice similarity coefficient and Hausdorff distance between the estimated-deep-network identified regions and the radiologist drawn annotations. We find the deep network architecture was able to segment the prostate gland anatomy with an average dice score of 0.85 in Test Cohort #1 (ProstateX, n=192, not part of training), 0.79 in Test Cohort #2 (TCIA Prostatectomy, n=26), 0.81 in Test Cohort #3 (TCIA Repeatability, n=15) and in Test Cohort #4 (Promise12, n=50) had an average dice score of 0.62. We also find the dice coefficient improved with larger prostate volumes in three of the four test cohorts. The variation of the dice scores from different cohorts of test images suggests the necessity of a universal model for prostate and PZ segmentation.