Automatic real-time prostate detection in transabdominal ultrasound images

Tiziano Natali, Mark Wijkhuizen, Liza Kurucz, Matteo Fusaglia, Pim J. van Leeuwen, Theo J.M. Ruers, Behdad Dashtbozorg

Keywords: CAD, Detection, Prostate, Transformer, CNN, Real-Time, Ultrasound imaging

Abstract:

Prostate cancer is the second most common form of cancer in men, thus easily accessible early diagnostic
tools are of vital importance. Transabdominal ultrasound is an inexpensive, non-invasive, and accessible
imaging modality. However, generated images are challenging to interpret and require expert clinicians
to be assessed. Therefore, we propose a DL model for real-time automatic detection of the prostate
for guidance to inexpert operators. Results show that the proposed model has similar performance
to state-of-the-art models, achieving mean average precision at 0.50 of 0.95 and queries per second of
993, indicating possible clinical application and possible improvement with more extensive pretraining
strategies.

Submission Number: 56