RESUMEN
BACKGROUND: The pancreas is a complex abdominal organ with many anatomical variations, and therefore automated pancreas segmentation from medical images is a challenging application. PURPOSE: In this paper, we present a framework for segmenting individual pancreatic subregions and the pancreatic duct from three-dimensional (3D) computed tomography (CT) images. METHODS: A multiagent reinforcement learning (RL) network was used to detect landmarks of the head, neck, body, and tail of the pancreas, and landmarks along the pancreatic duct in a selected target CT image. Using the landmark detection results, an atlas of pancreases was nonrigidly registered to the target image, resulting in anatomical probability maps for the pancreatic subregions and duct. The probability maps were augmented with multilabel 3D U-Net architectures to obtain the final segmentation results. RESULTS: To evaluate the performance of our proposed framework, we computed the Dice similarity coefficient (DSC) between the predicted and ground truth manual segmentations on a database of 82 CT images with manually segmented pancreatic subregions and 37 CT images with manually segmented pancreatic ducts. For the four pancreatic subregions, the mean DSC improved from 0.38, 0.44, and 0.39 with standard 3D U-Net, Attention U-Net, and shifted windowing (Swin) U-Net architectures, to 0.51, 0.47, and 0.49, respectively, when utilizing the proposed RL-based framework. For the pancreatic duct, the RL-based framework achieved a mean DSC of 0.70, significantly outperforming the standard approaches and existing methods on different datasets. CONCLUSIONS: The resulting accuracy of the proposed RL-based segmentation framework demonstrates an improvement against segmentation with standard U-Net architectures.
RESUMEN
BACKGROUND: High dose rate brachytherapy is commonly used in the treatment of prostate cancer. Treatment planning is often performed under transrectal ultrasound (US) guidance, but brachytherapy needles can be challenging to digitize due to the presence of poor US conspicuity and imaging artifacts. The plan accuracy and quality, however, are dependent on the proper visualization of the needles with millimeter accuracy. PURPOSE: This work describes a technique for generating a color overlay of needle locations atop the grayscale US image. Prototype devices were developed to produce vibrations in the brachytherapy needles that generate a high contrast color Doppler (CD) signal that highlights the needle locations with superior contrast and reduced artifacts. Denoted by the acronym color VISION (Vibrationally Induced Shimmering for Identifying an Object's Nature), the technology has the potential to improve applicator conspicuity and facilitate automated applicator digitization. METHODS: Three prototype vibrational devices with frequencies between 200-450 Hz were designed in-house and evaluated with needle implants in a phantom and cadaveric male pelvis using: (1) an actuator attached to the front of a prostate needle template; (2) an actuator attached to the top of the needle template; and (3) a hand-held actuator with a stylet, inserted directly into a needle's inner lumen. Acquired images were postprocessed in MATLAB to evaluate the potential for automated digitization. RESULTS: All prototype devices produced localized shimmering in implanted brachytherapy needles in both the axial and sagittal planes. The template mounted actuators provided better vibrational coupling and ease of operation than the stylet prototype. The Michelson contrast, or visibility, of the shimmering CD signal was 100% compared with ≤40% for B-mode imaging of a single needle. Proof-of-principle for automated applicator digitization using only the CD signal was demonstrated. CONCLUSIONS: The color VISION prototype devices successfully coupled mechanical vibrations into brachytherapy needles to generate US CD shimmering and accurately highlight brachytherapy needle locations. The high contrast and natively registered signal are promising for future work to automate the needle digitization and provide a real-time visual overlay of the applicator on the B-mode US image.