Combined ultrasound-fluoroscopy approach to the intraoperative detection of seeds in prostate brachytherapy Vishruta A. Dumane, Marco Zaider, Gilad N. Cohen and Michael F. Worman Department of Medical Physics, Memorial Sloan-Kettering Cancer Center, New York, NY One approach to seed detection in ultrasound is the use of parameters/features such as statistical moments, texture, gradient and median that can detect a seed and separate it from the background/surrounding tissue.  However, due to the inherent noisy nature of the ultrasonic images, 100% seed detection with no false positives within +/- 1-2mm of accuracy using parameters/features alone is a non-trivial problem. Since seed insertion is performed under fluoroscopic guidance, (where seeds are comparatively more visible), it may be possible to combine this information with features/parameters developed for localization of all the seeds in ultrasound with no false positives. Conclusions Detection of brachytherapy seeds in ultrasonic B-mode of the prostate may be accomplished through the use of parameters that can appropriately capture the acoustic signature of a seed in the image. Due to the inherent noisy nature of the images, the development of features that can detect all the seeds and yet offer no false positives is a challenging task. In order to reach 100% detection with no false positives, seeds in the ultrasound volume could be detected based on fluoroscopy images, where they are comparatively more visible. A combination of ultrasound and two fluoroscopic images appears to ensure 100% seed detection at 0% false positive rate and therefore could hold promise for intraoperative dynamic dosimetry. Introduction Intraoperative treatment planning for permanent prostate implants is recommended by the ABS. However, delivery of the planned dose is hindered by unavoidable fluctuations in seed positioning. Transrectal ultrasound image acquisition at 1 mm intervals Motivation There is a need to minimize the effects of discrepancies between the planned and the actual deposited seed locations intraoperatively, the prerequisite for which is a technique that can accurately and reliably localize the seeds. Ultrasound is an attractive imaging modality since treatment plans are generally based on its guidance. Transducer prostate seed single transverse scan Transverse B-mode image of a prostate phantom with detected seeds (white highlighted spots). Some false positives also appear. Materials and Methods Transverse B-mode image of a tissue-mimicking phantom (CIRS) containing 75 dummy seeds were acquired at 1 mm intervals at a center frequency of 6.5 MHz. Two fluoroscopy images of the phantom were obtained simultaneously during ultrasound scanning at +/- 15 degrees. Each seed on a given image corresponds to one and only one true seed on the second image as well as in the respective B-mode image. In order to determine the location of a given seed on an image in ultrasound, seed candidates on the second image that could be potential matches for this seed on the first image are identified and then backprojected onto the corresponding B-mode image in ultrasound. The intersections of the backprojected lines define possible regions where the given seed on the first image might exist. Parameters developed for seed detection are utilized to detect that particular seed in these regions. A 1 and B 1 are seeds on the first image (source position S 1 ) while A 2 and B 2 are the corresponding true matches on the second image (source position S 2 ). The backprojected points in the ultrasound volume are A, B, C and D, where the features are applied to detect seeds. Since seeds are physically present only at A and B, the features pick up seeds only in these regions A B S1S1 S2S2 A2A2 B2B2 A1A1 B1B1 C D (A) (B) Fluoroscopic images of the phantom with 75 seeds and the dummy probe inserted with 6 markers. Image (A) is taken at - 15° and image (B) is taken at +15° Results Given the positional information about where the actual seeds might exist in the corresponding ultrasonic B-mode images, parameters based on moments, texture analysis, gradient and median were utilized to detect the seeds within +/- 1-2 mm of the given position and also within +/- 3 mm intervals of the given position. A combination of transverse B-mode images and two fluoroscopy images could detect all the 75 seeds in the phantom within an accuracy of +/ mm with no false positives. Correspondence: