This paper proposes a robust and efficient needle detection method

which is used to localize and track the needle in three-dimensional (3D) transrectal ultrasound (TRUS) guided prostate therapy.First

all voxels are grouped into different line support regions (LSR) based on the outer product of adjacent voxels' gradient vectors.The needle axis is extracted by least square fitting in LSR.The needle endpoint is localized by finding an intensity drop along the needle axis.Evaluation results in synthetic data

tissue-mimicking agar

chicken breast phantoms and 3D TRUS patient images obtained during the prostate cryotherapy show that the proposed methods is with a relatively higher robustness and accuracy.The result of the in-vivo test also shows that our method outperformed several alternative methods in needle endpoint localization accuracy and TP rate.It is concluded that the proposed method is suitable for 3D TRUS guided prostate transperineal therapy.