KEYWORDS: Image segmentation, Education and training, Data modeling, Medical imaging, Performance modeling, Machine learning, Current controlled current source, Cardiovascular magnetic resonance imaging, Ablation, Heart
PurposeNeural networks have potential to automate medical image segmentation but require expensive labeling efforts. While methods have been proposed to reduce the labeling burden, most have not been thoroughly evaluated on large, clinical datasets or clinical tasks. We propose a method to train segmentation networks with limited labeled data and focus on thorough network evaluation.ApproachWe propose a semi-supervised method that leverages data augmentation, consistency regularization, and pseudolabeling and train four cardiac magnetic resonance (MR) segmentation networks. We evaluate the models on multiinstitutional, multiscanner, multidisease cardiac MR datasets using five cardiac functional biomarkers, which are compared to an expert’s measurements using Lin’s concordance correlation coefficient (CCC), the within-subject coefficient of variation (CV), and the Dice coefficient.ResultsThe semi-supervised networks achieve strong agreement using Lin’s CCC (>0.8), CV similar to an expert, and strong generalization performance. We compare the error modes of the semi-supervised networks against fully supervised networks. We evaluate semi-supervised model performance as a function of labeled training data and with different types of model supervision, showing that a model trained with 100 labeled image slices can achieve a Dice coefficient within 1.10% of a network trained with 16,000+ labeled image slices.ConclusionWe evaluate semi-supervision for medical image segmentation using heterogeneous datasets and clinical metrics. As methods for training models with little labeled data become more common, knowledge about how they perform on clinical tasks, how they fail, and how they perform with different amounts of labeled data is useful to model developers and users.
We report a novel photoacoustic (PA) scoring method for the risk stratification of thyroid nodules, which is combination of the American Thyroid Association (ATA) and PA malignancy probability. We performed multi-spectral PA imaging and multi-parametric PA analysis for thyroid cancer patients (23 malignancy and 29 benign cases). Initial multi-parametric PA analysis showed that malignancy of the thyroid nodules can be diagnosed with a 78% sensitivity and 93% specificity. Moreover, our novel score called ATAP improved the sensitivity to 83% while maintaining the specificity. The results suggest that the ATAP may help physicians examine thyroid nodules, thus reducing unnecessary biopsies.
Thyroid cancer is one of the most commonly diagnosed cancers in the world. Ultrasonography and fine-needle aspiration biopsy are the typical standard-of-care method for diagnosing thyroid nodules. However, about 20% of fine-needle aspiration biopsies generate undeterminable results, which can lead to overdiagnosis and overtreatment. In this study, we propose photoacoustic imaging as an additional triaging tool for identifying cancerous nodules in vivo. We enrolled and photoacoustically imaged 28 patients (19 malignant and 9 benign) who have thyroid nodules. To perform multispectral analysis, we used a series of 5 different wavelengths (i.e., 700, 756, 796, 866, and 900 nm), which were selected based on the optical absorption property of oxy- and deoxy-hemoglobin. All the raw data were automatically stored for further off-line processing, while the corresponding images were visualized on the clinical ultrasound machine in real-time. By using the multispectral photoacoustic data, we calculated the oxygen saturation values of the nodule areas. The result showed that the oxygen saturation level of malignant nodules was lower than that of benign nodules (p < 0.005), which matched with the well-known property of cancerous nodules. Based on the oxygen saturation value, malignant and benign nodules were differentiable with a sensitivity of 80% and specificity of 89%. The result showed the great potential of multispectral photoacoustic analysis as a novel method to identify malignancy of thyroid nodules in vivo. We also verified the robustness of the result by testing reproducibility and comparing inter-physician interpretation.
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 or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.