Deep learning segmentation of general interventional tools in two‐dimensional ultrasound images

Gillies, Derek J.; Rodgers, Jessica; Gyacskov, Igor; Roy, Priyanka; Kakani, Nirmal; Cool, Derek W.; Fenster, Aaron

doi:10.1002/mp.14427

Cited by 32 publications

(21 citation statements)

References 48 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…These combined analyses have the potential to enable the proposed method to overcome some of the limitations associated with previous methods that are based on B‐mode image analysis. For example, the methods by Gillies et al 26 . and Wijata et al., 27 which employed B‐mode image analysis to detect the needle, indicated an increase in the needle detection error when the needle is located near bright anatomical structures.…”

Section: Discussionmentioning

confidence: 99%

Needle detection using ultrasound B‐mode and power Doppler analyses

et al. 2022

View full text Add to dashboard Cite

Background: Ultrasound is employed in needle interventions to visualize the anatomical structures and track the needle. Nevertheless, needle detection in ultrasound images is a difficult task, specifically at steep insertion angles. Purpose: A new method is presented to enable effective needle detection using ultrasound B-mode and power Doppler analyses. Methods: A small buzzer is used to excite the needle and an ultrasound system is utilized to acquire B-mode and power Doppler images for the needle. The B-mode and power Doppler images are processed using Radon transform and local-phase analysis to initially detect the axis of the needle. The detection of the needle axis is improved by processing the power Doppler image using alpha shape analysis to define a region of interest (ROI) that contains the needle. Also, a set of feature maps is extracted from the ROI in the B-mode image. The feature maps are processed using a machine learning classifier to construct a likelihood image that visualizes the posterior needle likelihoods of the pixels. Radon transform is applied to the likelihood image to achieve an improved needle axis detection. Additionally, the region in the B-mode image surrounding the needle axis is analyzed to identify the needle tip using a custom-made probabilistic approach. Our method was utilized to detect needles inserted in ex vivo animal tissues at shallow [20 • − 40 • ), moderate [40 • − 60 • ), and steep [60 • − 85 • ] angles.Results: Our method detected the needles with failure rates equal to 0% and mean angle, axis, and tip errors less than or equal to 0.7 • , 0.6 mm, and 0.7 mm, respectively. Additionally, our method achieved favorable results compared to two recently introduced needle detection methods. Conclusions:The results indicate the potential of applying our method to achieve effective needle detection in ultrasound images.

show abstract

Section: Discussionmentioning

confidence: 99%

Needle detection using ultrasound B‐mode and power Doppler analyses

et al. 2022

View full text Add to dashboard Cite

show abstract

“…However, for this study, the breast phantom model sufficiently demonstrates the utility of the end‐effector biopsy device to align its needle axis to the RCM for targeted biopsy. To improve needle tracking and account for deviations in the tissue, the implementation of robust deep learning‐based needle segmentation methods in real‐time US images may improve targeted biopsy procedures 47 …”

Section: Discussionmentioning

confidence: 99%

“…To improve needle tracking and account for deviations in the tissue, the implementation of robust deep learning-based needle segmentation methods in real-time US images may improve targeted biopsy procedures. 47 Patient motion may provide another limitation to our proposed approach. In clinical practice, the breast will be gently immobilized and stabilized during PEM imaging and tissue sampling, mitigating the effect of motion and tissue deflection during the intervention.…”

Section: D Limitationsmentioning

confidence: 99%

Development of a mechatronic guidance system for targeted ultrasound‐guided biopsy under high‐resolution positron emission mammography localization

et al. 2021

Self Cite

View full text Add to dashboard Cite

Image-guided needle biopsy of small, detectable lesions is crucial for early-stage diagnosis, treatment planning, and management of breast cancer. High-resolution positron emission mammography (PEM) is a dedicated functional imaging modality that can detect breast cancer independent of breast tissue density, but anatomical context and real-time needle visualization are not yet available to guide biopsy. We propose a mechatronic guidance system integrating an ultrasound (US)-guided core-needle biopsy (CNB) with high-resolution PEM localization to improve the spatial sampling of breast lesions. This paper presents the benchtop testing and phantom studies to evaluate the accuracy of the system and its constituent components for targeted PEM-US-guided biopsy under simulated high-resolution PEM localization. Methods: A mechatronic guidance system was developed to operate with the Radialis PEM system and a conventional US system. The system includes a user-operated guidance arm and end-effector biopsy device, integrating a US transducer and CNB gun, with its needle focused on a remote center of motion (RCM). Custom software modules were developed to track, display, and guide the end-effector biopsy device. Registration of the mechatronic guidance system to a simulated PEM detector plate was performed using a landmark-based method. Testing was performed with fiducials positioned in the peripheral and central regions of the simulated detector plate and registration error was quantified. Breast phantom experiments were performed under ideal detection and localization to evaluate for bias in the end-effector biopsy device. The accuracy of the complete mechatronic guidance system to perform targeted breast biopsy was assessed using breast phantoms with simulated lesions. Three-dimensional positioning error was quantified, and principal component analysis assessed for directional trends in 3D space within 95% prediction intervals. Targeted breast biopsies with test phantoms were performed and an overall in-plane needle targeting error was quantified. Results: The mean registration errors were 0.63 mm (N = 44) and 0.73 mm (N = 72) in the peripheral and central regions of the simulated PEM detector plate, respectively. A 3D 95% prediction ellipsoid shows an error volume <2.0 mm in diameter, centered on the mean registration error. Under ideal detection and localization, targets <1.0 mm in diameter can be sampled with 95% confidence. The complete mechatronic guidance system was able to successfully spatially sample simulated breast lesions, 4 mm and 6 mm in diameter and height (N = 20) in known 3D positions in the PEM image coordinate space. The 3D positioning error was 0.85 mm (N = 20) with 0.64 mm in-plane and 0.44 mm cross-plane component errors. Targeted breast biopsies resulted in a mean in-plane needle targeting error of 1.08 mm (N = 15) allowing for targets 1.32 mm in radius to be sampled with 95% confidence. Conclusions: We demonstrated the utility of our mechatronic guidance system for targeted breast biopsy under high...

show abstract

“…However, it is challenging to automate these algorithms on US images with a variety of tissue backgrounds and needle contrasts. Deep learning (DL) based models especially convolutional neural networks have demonstrated competitive robustness and accuracy [18] , [19] , [20] , however, large clinical datasets with fine annotations are usually required for clinical applications but difficult to obtain.…”

Section: Introductionmentioning

confidence: 99%