Object detection as an aid for locating the prostate in surface-based abdominal ultrasound images

Automatic object detection is increasingly used in the medical field to enhance clinical workflows before, during, and after diagnosis of various conditions. One example is prostate detection and prostate volume estimation, which can aid in triaging patients for prostate cancer through risk-stratification using prostate-specific antigen density. In this paper, a baseline prostate detection framework is presented, highlighting that current state-of-the-art object detection models can detect the prostate in difficult to interpret surface-based ultrasound images. A 5-fold cross-validation study returned intersection-over-union, precision, recall, F1, and average-precision values above <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"><mml:mi>0.7</mml:mi></mml:math> with real-time capabilities possible. Additionally, a simple size calculation based on the detection results showed high correlation with ground truth measurements, with Pearson Correlation Coefficients ranging from <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"><mml:mi>0.55</mml:mi></mml:math> to <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"><mml:mi>0.84</mml:mi></mml:math> for prostate volume estimates. These findings will contribute to the development of a real-time prostate detection and size estimation platform for prostate cancer risk-stratification to reduce unnecessary biopsy rates in healthcare systems.