Proc. SPIE. 11513, 15th International Workshop on Breast Imaging (IWBI2020)
KEYWORDS: Breast, Point spread functions, Convolutional neural networks, Imaging systems, Computer simulations, Monte Carlo methods, Dual energy imaging, Digital breast tomosynthesis
Dual energy contrast-enhanced digital breast tomosynthesis (CEDBT) uses weighted subtraction of two energy spectra to highlight tumor angiogenesis with uptake of iodinated contrast agent. The high energy scan contains more severe scatter radiation than regular low energy DBT. The purpose of this study is to develop a convolutional neural network (CNN) based scatter correction method for dual energy CEDBT in both craniocaudal (CC) view and mediolateral oblique (MLO) view. Anthropomorphic digital breast phantoms with various glandularity and 3D shape were generated using the VICTRE software tool developed by the FDA. The pectoralis muscle layer was inserted into the phantoms for MLO view. Projection images with and without scatter radiation were simulated using Monte Carlo (MC) simulation code of VICTRE, meeting the prototype Siemens Mammomat Inspiration CEDBT system with 300 μm thick a-Se detector, 25 projections within 46-degree angular range. Scatter radiation ground truth was generated from MC simulated projection images to train CNN. Two separate U-net CNNs were trained to predict scatter radiation maps. Mean absolute percentage error (MAPE) was used as the loss function. The average MAPE of this method is less than 3 % from the ground truth of MC simulation. The proposed scatter correction method was then applied to clinical cases, demonstrating the reduction of cupping artifact and the improvement in contrast object conspicuity.
The image quality of contrast-enhanced digital breast tomosynthesis (CEDBT) is degraded by scatter radiation. Scatter correction can improve the object contrast and reduce the cupping artifacts, but the image quality is limited by the increased image noise. In this study we investigate the effect of scatter correction on image noise in CEDBT. A scatter correction method based on image convolution with scatter-to-primary ratio kernel was applied. We analyzed the noise power spectrum (NPS) for CEDBT projection images before and after scatter correction using CIRS breast phantoms and evaluated the signal-difference-to-noise ratio (SDNR) of the iodine objects after image reconstruction. We applied image filtering to reduce image noise after scatter correction for phantom and clinical images. A deep learning based denoising technique was applied to further reduce the image noise for clinical images. Our results show that the scatter correction increases the image noise in dual-energy subtracted images, and the improvement in SDNR from scatter correction is limited. Noise reduction applied after scatter removal can regain the benefit in SDNR from scatter correction and further improve the visualization of contrast enhancement in CEDBT.
Melanoma is the most dangerous form of skin cancer that often resembles moles. Dermatologists often recommend regular skin examination to identify and eliminate Melanoma in its early stages. To facilitate this process, we propose a hand-held computer (smart-phone, Raspberry Pi) based assistant that classifies with the dermatologist-level accuracy skin lesion images into malignant and benign and works in a standalone mobile device without requiring network connectivity. In this paper, we propose and implement a hybrid approach based on advanced deep learning model and domain-specific knowledge and features that dermatologists use for the inspection purpose to improve the accuracy of classification between benign and malignant skin lesions. Here, domain-specific features include the texture of the lesion boundary, the symmetry of the mole, and the boundary characteristics of the region of interest. We also obtain standard deep features from a pre-trained network optimized for mobile devices called Google's MobileNet. The experiments conducted on ISIC 2017 skin cancer classification challenge demonstrate the effectiveness and complementary nature of these hybrid features over the standard deep features. We performed experiments with the training, testing and validation data splits provided in the competition. Our method achieved area of 0.805 under the receiver operating characteristic curve. Our ultimate goal is to extend the trained model in a commercial hand-held mobile and sensor device such as Raspberry Pi and democratize the access to preventive health care.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print format on
SPIE.org.