PD56-11 RAW MICRO-ULTRASOUND TISSUE CHARACTERIZATION USING CONVOLUTION NEURAL NETWORKS TO DIFFERENTIATION BENIGN TISSUE FROM CLINICALLY SIGNIFICANT PROSTATE CANCER

You have accessJournal of UrologyProstate Cancer: Detection & Screening VI (PD56)1 Sep 2021PD56-11 RAW MICRO-ULTRASOUND TISSUE CHARACTERIZATION USING CONVOLUTION NEURAL NETWORKS TO DIFFERENTIATION BENIGN TISSUE FROM CLINICALLY SIGNIFICANT PROSTATE CANCER Ahmed El Kaffas, Brian Wodlinger, Mirabela Rusu, Aya Kamaya, Richard Fan, and Geoffrey Sonn Ahmed El KaffasAhmed El Kaffas More articles by this author , Brian WodlingerBrian Wodlinger More articles by this author , Mirabela RusuMirabela Rusu More articles by this author , Aya KamayaAya Kamaya More articles by this author , Richard FanRichard Fan More articles by this author , and Geoffrey SonnGeoffrey Sonn More articles by this author View All Author Informationhttps://doi.org/10.1097/JU.0000000000002090.11AboutPDF ToolsAdd to favoritesDownload CitationsTrack CitationsPermissionsReprints ShareFacebookLinked InTwitterEmail Abstract INTRODUCTION AND OBJECTIVE: Micro-ultrasound is a promising new technique offering high-resolution images to identify prostate cancer. However, clinical interpretation of the images remains challenging. The underlying raw radio-frequency (RF) ultrasound data acquired during imaging is a unique source of tissue acoustic properties, potentially ideal to identify suspicious areas to target at biopsy. Here, we develop a convolutional neural network (3D CNN) to classify tissue as benign vs clinically significant prostate cancer (csPCa) using the raw RF data. METHODS: 837 male patients (median age 63, IQR 57-68) undergoing biopsy for suspected prostate cancer were recruited. After IRB approval, data was acquired from 29MHz micro-ultrasound imaging (ExactVu, Exact Imaging, Markham, Canada) during targeted biopsy at 5 collaborating sites. Patients across all sites were scanned using consistent acquisition presets; all data was saved in raw digital RF format. One RF image was obtained at each biopsy location (up to 12 images per patient). Histopathological analysis of each biopsy sample served as the clinical standard of benign vs clinically significant cancer (grade group ≥2). Samples containing grade group 1 cancer were not analyzed. Data was processed to maintain both spatial and frequency information for images. A shallow 3D CNN model was trained on images (patient stratified); a similar network was also trained with the addition of PSA for each patient as a feature. Training used 5220 patient images. Testing was carried out on a set-aside dataset of 1310 images. The area under the receiver operator curve (ROC-AUC) was the main evaluation metric. We also explored whether our models could differentiate cancer from benign on a patient-level. RESULTS: Our model yielded a ROC-AUC of 82% in differentiating benign vs. csPSa at an image-level. Including the PSA into our model during training increased the ROC-AUC to 85%. On a patient-level, an 81% ROC-AUC was achieved in differentiating patients with benign vs. csPSa. Including PSA in the model increased the ROC-AUC to 87%. CONCLUSIONS: Our CNN model accurately differentiated cancerous from benign tissue using raw RF data obtained during clinical micro-ultrasound imaging. Source of Funding: NA © 2021 by American Urological Association Education and Research, Inc.FiguresReferencesRelatedDetails Volume 206Issue Supplement 3September 2021Page: e1008-e1008 Advertisement Copyright & Permissions© 2021 by American Urological Association Education and Research, Inc.MetricsAuthor Information Ahmed El Kaffas More articles by this author Brian Wodlinger More articles by this author Mirabela Rusu More articles by this author Aya Kamaya More articles by this author Richard Fan More articles by this author Geoffrey Sonn More articles by this author Expand All Advertisement Loading ...

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.