USFM: A universal ultrasound foundation model generalized to tasks and organs towards label efficient image analysis.

Inadequate generality across different organs and tasks constrains the application of ultrasound (US) image analysis methods in smart healthcare. Building a universal US foundation model holds the potential to address these issues. Nevertheless, the development of such foundation models encounters intrinsic challenges in US analysis, i.e., insufficient databases, low quality, and ineffective features. In this paper, we present a universal US foundation model, named USFM, generalized to diverse tasks and organs towards label efficient US image analysis. First, a large-scale Multi-organ, Multi-center, and Multi-device US database was built, comprehensively containing over two million US images. Organ-balanced sampling was employed for unbiased learning. Then, USFM is self-supervised pre-trained on the sufficient US database. To extract the effective features from low-quality US images, we proposed a spatial-frequency dual masked image modeling method. A productive spatial noise addition-recovery approach was designed to learn meaningful US information robustly, while a novel frequency band-stop masking learning approach was also employed to extract complex, implicit grayscale distribution and textural variations. Extensive experiments were conducted on the various tasks of segmentation, classification, and image enhancement from diverse organs and diseases. Comparisons with representative US image analysis models illustrate the universality and effectiveness of USFM. The label efficiency experiments suggest the USFM obtains robust performance with only 20% annotation, laying the groundwork for the rapid development of US models in clinical practices.

Standardization of ultrasound images across various centers: M2O-DiffGAN bridging the gaps among unpaired multi-domain ultrasound images.

Domain shift problem is commonplace for ultrasound image analysis due to difference imaging setting and diverse medical centers, which lead to poor generalizability of deep learning-based methods. Multi-Source Domain Transformation (MSDT) provides a promising way to tackle the performance degeneration caused by the domain shift, which is more practical and challenging compared to conventional single-source transformation tasks. An effective unsupervised domain combination strategy is highly required to handle multiple domains without annotations. Fidelity and quality of generated images are also important to ensure the accuracy of computer-aided diagnosis. However, existing MSDT approaches underperform in above two areas. In this paper, an efficient domain transformation model named M2O-DiffGAN is introduced to achieve a unified mapping from multiple unlabeled source domains to the target domain. A cycle-consistent "many-to-one" adversarial learning architecture is introduced to model various unlabeled domains jointly. A condition adversarial diffusion process is employed to generate images with high-fidelity, combining an adversarial projector to capture reverse transition probabilities over large step sizes for accelerating sampling. Considering the limited perceptual information of ultrasound images, an ultrasound-specific content loss helps to capture more perceptual features for synthesizing high-quality ultrasound images. Massive comparisons on six clinical datasets covering thyroid, carotid and breast demonstrate the superiority of the M2O-DiffGAN in the performance of bridging the domain gaps and enlarging the generalization of downstream analysis methods compared to state-of-the-art algorithms. It improves the mean MI, Bhattacharyya Coefficient, dice and IoU assessments by 0.390, 0.120, 0.245 and 0.250, presenting promising clinical applications.

Weakly Supervised Lesion Detection and Diagnosis for Breast Cancers with Partially Annotated Ultrasound Images.

Deep learning (DL) has proven highly effective for ultrasound-based computer-aided diagnosis (CAD) of breast cancers. In an automatic CAD system, lesion detection is critical for the following diagnosis. However, existing DL-based methods generally require voluminous manually-annotated region of interest (ROI) labels and class labels to train both the lesion detection and diagnosis models. In clinical practice, the ROI labels, i.e. ground truths, may not always be optimal for the classification task due to individual experience of sonologists, resulting in the issue of coarse annotation to limit the diagnosis performance of a CAD model. To address this issue, a novel Two-Stage Detection and Diagnosis Network (TSDDNet) is proposed based on weakly supervised learning to improve diagnostic accuracy of the ultrasound-based CAD for breast cancers. In particular, all the initial ROI-level labels are considered as coarse annotations before model training. In the first training stage, a candidate selection mechanism is then designed to refine manual ROIs in the fully annotated images and generate accurate pseudo-ROIs for the partially annotated images under the guidance of class labels. The training set is updated with more accurate ROI labels for the second training stage. A fusion network is developed to integrate detection network and classification network into a unified end-to-end framework as the final CAD model in the second training stage. A self-distillation strategy is designed on this model for joint optimization to further improves its diagnosis performance. The proposed TSDDNet is evaluated on three B-mode ultrasound datasets, and the experimental results indicate that it achieves the best performance on both lesion detection and diagnosis tasks, suggesting promising application potential.

IMAGGS: a radiogenomic framework for identifying multi-way associations in breast cancer subtypes.

Investigating correlations between radiomic and genomic profiling in breast cancer (BC) molecular subtypes is crucial for understanding disease mechanisms and providing personalized treatment. We present a well-designed radiogenomic framework image-gene-gene set (IMAGGS), which detects multi-way associations in BC subtypes by integrating radiomic and genomic features. Our dataset consists of 721 patients, each of whom has 12 ultrasound (US) images captured from different angles and gene mutation data. To better characterize tumor traits, 12 multi-angle US images are fused using two distinct strategies. Then, we analyze complex many-to-many associations between phenotypic and genotypic features using a machine learning algorithm, deviating from the prevalent one-to-one relationship pattern observed in previous studies. Key radiomic and genomic features are screened using these associations. In addition, gene set enrichment analysis is performed to investigate the joint effects of gene sets and delve deeper into the biological functions of BC subtypes. We further validate the feasibility of IMAGGS in a glioblastoma multiforme dataset to demonstrate the scalability of IMAGGS across different modalities and diseases. Taken together, IMAGGS provides a comprehensive characterization for diseases by associating imaging, genes, and gene sets, paving the way for biological interpretation of radiomics and development of targeted therapy.

Application of ultrasound-based radiomics models of breast masses to predict invasive components of encapsulated papillary carcinoma.

Background: Axillary lymph node (ALN) metastasis is seen in encapsulated papillary carcinoma (EPC), mostly with an invasive component (INV). Radiomics can offer more information beyond subjective grayscale and color Doppler ultrasound (US) image interpretation. This study aimed to develop radiomics models for predicting an INV of EPC in the breast based on US images. Methods: This study retrospectively enrolled 105 patients (107 masses) with a pathological diagnosis of EPC from January 2016 to April 2021, and all masses had preoperative US images. Of the 107 masses, 64 were randomized to a training set and 43 to a test set. US and clinical features were analyzed to identify features associated with INVs. Then, based on the manually segmented US images to obtain radiomics features, the models to predict INVs were built with 5 ensemble machine learning classifiers. We estimated the performance of the predictive models using accuracy, the area under the receiver operating characteristic (ROC) curve (AUC), sensitivity, and specificity. Results: The mean age was 63.71 years (range, 31 to 85 years); the mean size of tumors was 23.40 mm (range, 9 to 120 mm). Among all clinical and US features, only shape was statistically different between EPC with INVs and those without (P<0.05). In this study, the models based on Random Under Sampling (RUS) Boost, Random Forest, XGBoost, AdaBoost, and Easy Ensemble methods had good performance, among which RUS Boost had the best performance with an AUC of 0.875 [95% confidence interval (CI): 0.750-0.974] in the test set. Conclusions: Radiomics prediction models were effective in predicting the INV of EPC, whereas clinical and US features demonstrated relatively decreased predictive utility.

An autocatalytic multicomponent DNAzyme nanomachine for tumor-specific photothermal therapy sensitization in pancreatic cancer.

Multicomponent deoxyribozymes (MNAzymes) have great potential in gene therapy, but their ability to recognize disease tissue and further achieve synergistic gene regulation has rarely been studied. Herein, Arginylglycylaspartic acid (RGD)-modified Distearyl acylphosphatidyl ethanolamine (DSPE)-polyethylene glycol (PEG) (DSPE-PEG-RGD) micelle is prepared with a DSPE hydrophobic core to load the photothermal therapy (PTT) dye IR780 and the calcium efflux pump inhibitor curcumin. Then, the MNAzyme is distributed into the hydrophilic PEG layer and sealed with calcium phosphate through biomineralization. Moreover, RGD is attached to the outer tail of PEG for tumor targeting. The constructed nanomachine can release MNAzyme and the cofactor Ca2+ under acidic conditions and self-assemble into an active mode to cleave heat shock protein (HSP) mRNA by consuming the oncogene miRNA-21. Silencing miRNA-21 enhances the expression of the tumor suppressor gene PTEN, leading to PTT sensitization. Meanwhile, curcumin maintains high intracellular Ca2+ to further suppress HSP-chaperone ATP by disrupting mitochondrial Ca2+ homeostasis. Therefore, pancreatic cancer is triple-sensitized to IR780-mediated PTT. The in vitro and in vivo results show that the MNAzyme-based nanomachine can strongly regulate HSP and PTEN expression and lead to significant pancreatic tumor inhibition under laser irradiation.

Unraveling the Pivotal Network of Ultrasound and Somatic Mutations in Triple-Negative and Non-Triple-Negative Breast Cancer.

Purpose: The emergence of genomic targeted therapy has improved the prospects of treatment for breast cancer (BC). However, genetic testing relies on invasive and sophisticated procedures. Patients and Methods: Here, we performed ultrasound (US) and target sequencing to unravel the possible association between US radiomics features and somatic mutations in TNBC (n=83) and non-TNBC (n=83) patients. Least absolute shrinkage and selection operator (Lasso) were utilized to perform radiomic feature selection. The Kyoto Encyclopedia of Genes and Genomes (KEGG) analysis was utilized to identify the signaling pathways associated with radiomic features. Results: Thirteen differently represented radiomic features were identified in TNBC and non-TNBC, including tumor shape, textual, and intensity features. The US radiomic-gene pairs were differently exhibited between TNBC and non-TNBC. Further investigation with KEGG verified radiomic-pathway (ie, JAK-STAT, MAPK, Ras, Wnt, microRNAs in cancer, PI3K-Akt) associations in TNBC and non-TNBC. Conclusion: The pivotal network provided the connections of US radiogenomic signature and target sequencing for non-invasive genetic assessment of precise BC treatment.

HAL-IA: A Hybrid Active Learning framework using Interactive Annotation for medical image segmentation.

High performance of deep learning models on medical image segmentation greatly relies on large amount of pixel-wise annotated data, yet annotations are costly to collect. How to obtain high accuracy segmentation labels of medical images with limited cost (e.g. time) becomes an urgent problem. Active learning can reduce the annotation cost of image segmentation, but it faces three challenges: the cold start problem, an effective sample selection strategy for segmentation task and the burden of manual annotation. In this work, we propose a Hybrid Active Learning framework using Interactive Annotation (HAL-IA) for medical image segmentation, which reduces the annotation cost both in decreasing the amount of the annotated images and simplifying the annotation process. Specifically, we propose a novel hybrid sample selection strategy to select the most valuable samples for segmentation model performance improvement. This strategy combines pixel entropy, regional consistency and image diversity to ensure that the selected samples have high uncertainty and diversity. In addition, we propose a warm-start initialization strategy to build the initial annotated dataset to avoid the cold-start problem. To simplify the manual annotation process, we propose an interactive annotation module with suggested superpixels to obtain pixel-wise label with several clicks. We validate our proposed framework with extensive segmentation experiments on four medical image datasets. Experimental results showed that the proposed framework achieves high accuracy pixel-wise annotations and models with less labeled data and fewer interactions, outperforming other state-of-the-art methods. Our method can help physicians efficiently obtain accurate medical image segmentation results for clinical analysis and diagnosis.

Deep learning radiomics model based on breast ultrasound video to predict HER2 expression status.

Purpose: The detection of human epidermal growth factor receptor 2 (HER2) expression status is essential to determining the chemotherapy regimen for breast cancer patients and to improving their prognosis. We developed a deep learning radiomics (DLR) model combining time-frequency domain features of ultrasound (US) video of breast lesions with clinical parameters for predicting HER2 expression status. Patients and Methods: Data for this research was obtained from 807 breast cancer patients who visited from February 2019 to July 2020. Ultimately, 445 patients were included in the study. Pre-operative breast ultrasound examination videos were collected and split into a training set and a test set. Building a training set of DLR models combining time-frequency domain features and clinical features of ultrasound video of breast lesions based on the training set data to predict HER2 expression status. Test the performance of the model using test set data. The final models integrated with different classifiers are compared, and the best performing model is finally selected. Results: The best diagnostic performance in predicting HER2 expression status is provided by an Extreme Gradient Boosting (XGBoost)-based time-frequency domain feature classifier combined with a logistic regression (LR)-based clinical parameter classifier of clinical parameters combined DLR, particularly with a high specificity of 0.917. The area under the receiver operating characteristic curve (AUC) for the test cohort was 0.810. Conclusion: Our study provides a non-invasive imaging biomarker to predict HER2 expression status in breast cancer patients.

An omics-to-omics joint knowledge association subtensor model for radiogenomics cross-modal modules from genomics and ultrasonic images of breast cancers.

The radiogenomics analysis can provide the connections between genomics and radiomics, which can infer the genomic features of tumors from their radiogenomic associations through the low-cost and non-invasiveness screening ultrasonic images. Although there are a number of pioneer approaches exploring the connections between genomic aberrations and ultrasonic features, these studies mainly focus on the relationship between ultrasonic features and only the most popular cancer genes, confronting two difficulties: missing many-to-many relationships as omics-to-omics view, and confounding group-specific associations with whole sample associations. To overcome the difficulty of omics-to-omics view and the issue of tumor heterogeneity, we propose an omics-to-omics joint knowledge association subtensor model. Specifically, the subtensor factorization framework can successfully discover the joint cross-modal module via an omics-to-omics view, while the sparse weight sample indication strategy can mine sample subgroups from the multi-omic data with tumor heterogeneity. The experimental evaluation result shows the jointness of the discovered modules across omics, their association with tumorigenesis contribution, and their relation for cancer related functions. In summary, our proposed omics-to-omics joint knowledge association subtensor model can serve as an efficient tool for radiogenomic knowledge associations, promoting the cross-modal knowledge graph construction of in explainable artificial intelligence cancer diagnosis.

Biomimetic nanoparticles drive the mechanism understanding of shear-wave elasticity stiffness in triple negative breast cancers to predict clinical treatment.

In clinical practice, we noticed that triple negative breast cancer (TNBC) patients had higher shear-wave elasticity (SWE) stiffness than non-TNBC patients and a higher α-SMA expression was found in TNBC tissues than the non-TNBC tissues. Moreover, SWE stiffness also shows a clear correlation to neoadjuvant response efficiency. To elaborate this phenomenon, TNBC cell membrane-modified polylactide acid-glycolic acid (PLGA) nanoparticle was fabricated to specifically deliver artesunate to regulate SWE stiffness through inhibiting CAFs functional status. As tested in MDA-MB-231 and E0771 orthotopic tumor models, CAFs functional status inhibited by 231M-ARS@PLGA nanoparticles (231M-AP NPs) had reduced the SWE stiffness as well as attenuated hypoxia of tumor as tumor soil loosening agent which amplified the antitumor effects of paclitaxel and PD1 inhibitor. Single-cell sequencing indicated that the two main CAFs (extracellular matrix and wound healing CAFs) that produces extracellular matrix could influence the tumor SWE stiffness as well as the antitumor effect of drugs. Further, biomimetic nanoparticles inhibited CAFs function could attenuate tumor hypoxia by increasing proportion of inflammatory blood vessels and oxygen transport capacity. Therefore, our finding is fundamental for understanding the role of CAFs on affecting SWE stiffness and drugs antitumor effects, which can be further implied in the potential clinical theranostic predicting in neoadjuvant therapy efficacy through non-invasive analyzing of SWE imaging.

The outcomes and prognostic factors of patients who underwent reoperation for persistent/recurrent papillary thyroid carcinoma.

BACKGROUND: While the most suitable approach for treating persistent/recurrent papillary thyroid carcinoma (PTC) remains controversial, reoperation may be considered an effective method. The efficacy of reoperation in patients with locoregional persistent/recurrent PTC, especially those with unsatisfactory radioactive iodine (RAI) ablation results, is still uncertain. This study aimed to clarify the clinical management strategies for locoregional persistent/recurrent PTC and to explore factors that may affect long-term patient outcomes after reoperation. METHODS: In total, 124 patients who initially underwent thyroidectomy and variable extents of RAI therapy and finally received reoperation for locoregionally persistent/recurrent PTC were included. The parameters associated with recurrence-free survival (RFS) were analysed using a Cox proportional hazards model. RESULTS: Overall, 124 patients presented with structural disease after initial therapy and underwent secondary surgical resection, of whom 32 patients developed further structural disease during follow-up after reoperation. At the time of reoperation, metastatic lymph nodes with extranodal extension (P = 0.023) and high unstimulated thyroglobulin (unstim-Tg) levels after reoperation (post-reop) (P = 0.001) were independent prognostic factors for RFS. Neither RAI avidity nor the frequency and dose of RAI therapies before reoperation affected RFS. CONCLUSIONS: Reoperation is an ideal clinical treatment strategy for structural locoregional persistent/recurrent PTC, and repeated empirical RAI therapies performed prior to reoperation may not contribute to the long-term outcomes of persistent/recurrent PTC patients. Metastatic lymph nodes with extranodal extension and post-reop unstim-Tg > 10.1 ng/mL may predict a poor prognosis.

Joint Localization and Classification of Breast Cancer in B-Mode Ultrasound Imaging via Collaborative Learning With Elastography.

Convolutional neural networks (CNNs) have been successfully applied in the computer-aided ultrasound diagnosis for breast cancer. Up to now, several CNN-based methods have been proposed. However, most of them consider tumor localization and classification as two separate steps, rather than performing them simultaneously. Besides, they suffer from the limited diagnosis information in the B-mode ultrasound (BUS) images. In this study, we develop a novel network ResNet-GAP that incorporates both localization and classification into a unified procedure. To enhance the performance of ResNet-GAP, we leverage stiffness information in the elastography ultrasound (EUS) modality by collaborative learning in the training stage. Specifically, a dual-channel ResNet-GAP network is developed, one channel for BUS and the other for EUS. In each channel, multiple class activity maps (CAMs) are generated using a series of convolutional kernels of different sizes. The multi-scale consistency of the CAMs in both channels are further considered in network optimization. Experiments on 264 patients in this study show that the newly developed ResNet-GAP achieves an accuracy of 88.6%, a sensitivity of 95.3%, a specificity of 84.6%, and an AUC of 93.6% on the classification task, and a 1.0NLF of 87.9% on the localization task, which is better than some state-of-the-art approaches.

Ultrasound-based radiomics analysis for preoperative prediction of central and lateral cervical lymph node metastasis in papillary thyroid carcinoma: a multi-institutional study.

BACKGROUND: An accurate preoperative assessment of cervical lymph node metastasis (LNM) is important for choosing an optimal therapeutic strategy for papillary thyroid carcinoma (PTC) patients. This study aimed to develop and validate two ultrasound (US) nomograms for the individual prediction of central and lateral compartment LNM in patients with PTC. METHODS: A total of 720 PTC patients from 3 institutions were enrolled in this study. They were categorized into a primary cohort, an internal validation, and two external validation cohorts. Radiomics features were extracted from conventional US images. LASSO regression was used to select optimized features to construct the radiomics signature. Two nomograms integrating independent clinical variables and radiomics signature were established with multivariate logistic regression. The performance of the nomograms was assessed with regard to discrimination, calibration, and clinical usefulness. RESULTS: The radiomics scores were significantly higher in patients with central/lateral LNM. A radiomics nomogram indicated good discrimination for central compartment LNM, with an area under the curve (AUC) of 0.875 in the training set, the corresponding value in the validation sets were 0.856, 0.870 and 0.870, respectively. Another nomogram for predicting lateral LNM also demonstrated good performance with an AUC of 0.938 and 0.905 in the training and internal validation cohorts, respectively. The AUC for the two external validation cohorts were 0.881 and 0.903, respectively. The clinical utility of the nomograms was confirmed by the decision curve analysis. CONCLUSION: The nomograms proposed here have favorable performance for preoperatively predicting cervical LNM, hold promise for optimizing the personalized treatment, and might greatly facilitate the decision-making in clinical practice.

High Expression of THBS1 Leads to a Poor Prognosis in Papillary Thyroid Cancer and Suppresses the Anti-Tumor Immune Microenvironment.

Objectives: To study the role of thrombospondin-1 (THBS1) in papillary thyroid cancer (PTC) prognosis and the immune microenvironment. Methods: A retrospective cohort study was designed, and data from The Cancer Genome Atlas database and PTC tissues from Fudan University Shanghai Cancer Center were used. Weighted gene co-expression network analysis was performed to build a THBS1-immune-related gene prognostic index (T-I index). Results: High THBS1 expression was correlated with advanced TNM stage, higher recurrence risk, and shorter progression-free interval. High THBS1 expression correlated with MAPK and PD1 pathways indicating a tumor promoting and immunity-inhibiting tendency. The T-I index showed a powerful capacity to predict progression-free survival and immunotherapy benefit. Conclusion: High expression of THBS1 leads to a poor prognosis in PTCs and suppresses the anti-tumor immune microenvironment.

Breast Tumor Classification Based on MRI-US Images by Disentangling Modality Features.

Dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) and ultrasound (US), which are two common modalities for clinical breast tumor diagnosis besides Mammograms, can provide different and complementary information for the same tumor regions. Although many machine learning methods have been proposed for breast tumor classification based on either single modality, it remains unclear how to further boost the classification performance by utilizing paired multi-modality information with different dimensions. In this paper, we propose MRI-US multi-modality network (MUM-Net) to classify breast tumor into different subtypes based on 3D MR and 2D US images. The key insight of MUM-Net is that we explicitly distill modality-agnostic features for tumor classification. Specifically, we first adopt a discrimination-adaption module to decompose features into modality-agnostic and modality-specific ones with min-max training strategies. Then, we propose a feature fusion module to increase the compactness of the modality-agnostic features by utilizing an affinity matrix with nearest neighbour selection. We build a paired MRI-US breast tumor classification dataset containing 502 cases with three clinical indicators to validate the proposed method. In three tasks including lymph node metastasis, histological grade and Ki-67 level, MUM-Net achieves AUC scores of 0.8581, 0.8965 and 0.8577, outperforming other counterparts which are based on single task or single modality by a wide margin. In addition, we find that the extracted modality-agnostic features can help the network focus on the tumor regions in both modalities.

Application of preoperative ultrasound features combined with clinical factors in predicting HER2-positive subtype (non-luminal) breast cancer.

BACKGROUND: Human epidermal growth factor receptor2+ subtype breast cancer has a high degree of malignancy and a poor prognosis. The aim of this study is to develop a prediction model for the human epidermal growth factor receptor2+ subtype (non-luminal) of breast cancer based on the clinical and ultrasound features related with estrogen receptor, progesterone receptor, and human epidermal growth factor receptor2. METHODS: We collected clinical data and reviewed preoperative ultrasound images of enrolled breast cancers from September 2017 to August 2020. We divided the data into in three groups as follows. Group I: estrogen receptor ± , Group II: progesterone receptor ± and Group III: human epidermal growth factor receptor2 ± . Univariate and multivariate logistic regression analyses were used to analyze the clinical and ultrasound features related with biomarkers among these groups. A model to predict human epidermal growth factor receptor2+ subtype was then developed based on the results of multivariate regression analyses, and the efficacy was evaluated using the area under receiver operating characteristic curve, accuracy, sensitivity, specificity. RESULTS: The human epidermal growth factor receptor2+ subtype accounted for 138 cases (11.8%) in the training set and 51 cases (10.1%) in the test set. In the multivariate regression analysis, age ≤ 50 years was an independent predictor of progesterone receptor + (p = 0.007), and posterior enhancement was a negative predictor of progesterone receptor + (p = 0.013) in Group II; palpable axillary lymph node, round, irregular shape and calcifications were independent predictors of the positivity for human epidermal growth factor receptor-2 in Group III (p = 0.001, p = 0.007, p = 0.010, p < 0.001, respectively). In Group I, shape was the only factor related to estrogen receptor status in the univariate analysis (p < 0.05). The area under receiver operating characteristic curve, accuracy, sensitivity, specificity of the model to predict human epidermal growth factor receptor2+ subtype breast cancer was 0.697, 60.14%, 72.46%, 58.49% and 0.725, 72.06%, 64.71%, 72.89% in the training and test sets, respectively. CONCLUSIONS: Our study established a model to predict the human epidermal growth factor receptor2-positive subtype with moderate performance. And the results demonstrated that clinical and ultrasound features were significantly associated with biomarkers.

Radiogenomic Analysis of Papillary Thyroid Carcinoma for Prediction of Cervical Lymph Node Metastasis: A Preliminary Study.

BACKGROUND: Papillary thyroid carcinoma (PTC) is characterized by frequent metastases to cervical lymph nodes (CLNs), and the presence of lymph node metastasis at diagnosis has a significant impact on the surgical approach. Therefore, we established a radiomic signature to predict the CLN status of PTC patients using preoperative thyroid ultrasound, and investigated the association between the radiomic features and underlying molecular characteristics of PTC tumors. METHODS: In total, 270 patients were enrolled in this prospective study, and radiomic features were extracted according to multiple guidelines. A radiomic signature was built with selected features in the training cohort and validated in the validation cohort. The total protein extracted from tumor samples was analyzed with LC/MS and iTRAQ technology. Gene modules acquired by clustering were chosen for their diagnostic significance. A radiogenomic map linking radiomic features to gene modules was constructed with the Spearman correlation matrix. Genes in modules related to metastasis were extracted for Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analyses, and a protein-protein interaction (PPI) network was built to identify the hub genes in the modules. Finally, the screened hub genes were validated by immunohistochemistry analysis. RESULTS: The radiomic signature showed good performance for predicting CLN status in training and validation cohorts, with area under curve of 0.873 and 0.831 respectively. A radiogenomic map was created with nine significant correlations between radiomic features and gene modules, and two of them had higher correlation coefficient. Among these, MEmeganta representing the upregulation of telomere maintenance via telomerase and cell-cell adhesion was correlated with 'Rectlike' and 'deviation ratio of tumor tissue and normal thyroid gland' which reflect the margin and the internal echogenicity of the tumor, respectively. MEblue capturing cell-cell adhesion and glycolysis was associated with feature 'minimum calcification area' which measures the punctate calcification. The hub genes of the two modules were identified by protein-protein interaction network. Immunohistochemistry validated that LAMC1 and THBS1 were differently expressed in metastatic and non-metastatic tissues (p=0.003; p=0.002). And LAMC1 was associated with feature 'Rectlike' and 'deviation ratio of tumor and normal thyroid gland' (p<0.001; p<0.001); THBS1 was correlated with 'minimum calcification area' (p<0.001). CONCLUSIONS: The radiomic signature proposed here has the potential to noninvasively predict the CLN status in PTC patients. Merging imaging phenotypes with genomic data could allow noninvasive identification of the molecular properties of PTC tumors, which might support clinical decision making and personalized management.

Prediction of treatment responses to neoadjuvant chemotherapy in breast cancer using contrast-enhanced ultrasound.

BACKGROUND: Elucidation the efficacy of neoadjuvant chemotherapy (NAC) in breast cancer is important for informing therapeutic decisions. This study aimed at evaluating the potential value of contrast-enhanced ultrasound (CEUS) parameters in predicting breast cancer responses to NAC. METHODS: We performed CEUS examinations before and after two cycles of NAC. Quantitative CEUS parameters [maximum intensity (IMAX), rise time (RT), time to peak (TTP), and mean transit time (mTT)], tumor diameter, and their changes were measured and compared to histopathological responses, according to the Miller-Payne Grading (MPG) system (score 1, 2, or 3: minor response; score 4 or 5: good response). Prediction models for good response were developed by multiple logistic regression analysis and internally validated through bootstrap analysis. The receiver operating characteristic (ROC) curve was used to evaluate the performance of prediction models. RESULTS: A total of 143 patients were enrolled in this study among whom 98 (68.5%) achieved a good response and while 45 (31.5%) exhibited a minor response. Several imaging variables including diameter, IMAX, changes in diameter (Δdiameter), IMAX (ΔIMAX) and TTP (ΔTTP) were found to be significantly associated with good therapeutic responses (P<0.05). The areas under the curve (AUC) increased from 0.748 to 0.841 in the multivariate model that combined CEUS parameters and molecular subtypes with a sensitivity and specificity of 0.786, 0.745, respectively. Tumor molecular subtype was the primary predictor of primary endpoint. CONCLUSIONS: CEUS is a potential tool for predicting responses to NAC in locally advanced breast cancer patients. Compared to the other molecular subtypes, triple negative and HER2+/ER- subtypes are more likely to exhibit a good response to NAC.

Application of deep learning to predict underestimation in ductal carcinoma in situ of the breast with ultrasound.

BACKGROUND: To develop an ultrasound-based deep learning model to predict postoperative upgrading of pure ductal carcinoma in situ (DCIS) diagnosed by core needle biopsy (CNB) before surgery. METHODS: Of the 360 patients with DCIS diagnosed by CNB and identified retrospectively, 180 had lesions upstaged to ductal carcinoma in situ with microinvasion (DCISM) or invasive ductal carcinoma (IDC) postoperatively. Ultrasound images obtained from the hospital database were divided into a training set (n=240) and validation set (n=120), with a ratio of 2:1 in chronological order. Four deep learning models, based on the ResNet and VggNet structures, were established to classify the ultrasound images into postoperative upgrade and pure DCIS. We obtained the area under the receiver operating characteristic curve (AUROC), specificity, sensitivity, accuracy, positive predictive value (PPV), and negative predictive value (NPV) to estimate the performance of the predictive models. The robustness of the models was evaluated by a 3-fold cross-validation. RESULTS: Clinical features were not significantly different between the training set and the test set (P value >0.05). The area under the receiver operating characteristic curve of our models ranged from 0.724 to 0.804. The sensitivity, specificity, and accuracy of the optimal model were 0.733, 0.750, and 0.742, respectively. The three-fold cross-validation results showed that the model was very robust. CONCLUSIONS: The ultrasound-based deep learning prediction model is effective in predicting DCIS that will be upgraded postoperatively.