Active Learning in Brain Tumor Segmentation with Uncertainty Sampling and Annotation Redundancy Restriction.

Deep learning models have demonstrated great potential in medical imaging but are limited by the expensive, large volume of annotations required. To address this, we compared different active learning strategies by training models on subsets of the most informative images using real-world clinical datasets for brain tumor segmentation and proposing a framework that minimizes the data needed while maintaining performance. Then, 638 multi-institutional brain tumor magnetic resonance imaging scans were used to train three-dimensional U-net models and compare active learning strategies. Uncertainty estimation techniques including Bayesian estimation with dropout, bootstrapping, and margins sampling were compared to random query. Strategies to avoid annotating similar images were also considered. We determined the minimum data necessary to achieve performance equivalent to the model trained on the full dataset (α = 0.05). Bayesian approximation with dropout at training and testing showed results equivalent to that of the full data model (target) with around 30% of the training data needed by random query to achieve target performance (p = 0.018). Annotation redundancy restriction techniques can reduce the training data needed by random query to achieve target performance by 20%. We investigated various active learning strategies to minimize the annotation burden for three-dimensional brain tumor segmentation. Dropout uncertainty estimation achieved target performance with the least annotated data.

Head and neck tumor segmentation convolutional neural network robust to missing PET/CT modalities using channel dropout.

Objective. Radiation therapy for head and neck (H&N) cancer relies on accurate segmentation of the primary tumor. A robust, accurate, and automated gross tumor volume segmentation method is warranted for H&N cancer therapeutic management. The purpose of this study is to develop a novel deep learning segmentation model for H&N cancer based on independent and combined CT and FDG-PET modalities.Approach. In this study, we developed a robust deep learning-based model leveraging information from both CT and PET. We implemented a 3D U-Net architecture with 5 levels of encoding and decoding, computing model loss through deep supervision. We used a channel dropout technique to emulate different combinations of input modalities. This technique prevents potential performance issues when only one modality is available, increasing model robustness. We implemented ensemble modeling by combining two types of convolutions with differing receptive fields, conventional and dilated, to improve capture of both fine details and global information.Main Results. Our proposed methods yielded promising results, with a Dice similarity coefficient (DSC) of 0.802 when deployed on combined CT and PET, DSC of 0.610 when deployed on CT, and DSC of 0.750 when deployed on PET.Significance. Application of a channel dropout method allowed for a single model to achieve high performance when deployed on either single modality images (CT or PET) or combined modality images (CT and PET). The presented segmentation techniques are clinically relevant to applications where images from a certain modality might not always be available.

Surface-Based Falff: A Potential Novel Biomarker for Prediction of Radiation Encephalopathy in Patients With Nasopharyngeal Carcinoma.

Radiation encephalopathy (RE) is an important potential complication in patients with nasopharyngeal carcinoma (NPC) who undergo radiotherapy (RT) that can affect the quality of life. However, a functional imaging biomarker of pre-symptomatic RE has not yet been established. This study aimed to assess radiation-induced gray matter functional alterations and explore fractional amplitude of low-frequency fluctuation (fALFF) as an imaging biomarker for predicting or diagnosing RE in patients with NPC. A total of 60 patients with NPC were examined, 21 in the pre-RT cohort and 39 in the post-RT cohort. Patients in the post-RT cohort were further divided into two subgroups according to the occurrence of RE in follow-up: post-RT non-RE (n = 21) and post-RT REproved infollow-up (n = 18). Surface-based and volume-based fALFF were used to detect radiation-induced functional alterations. Functional derived features were then adopted to construct a predictive model for the diagnosis of RE. We observed that surface-based fALFF could sensitively detect radiation-induced functional alterations in the intratemporal brain regions (such as the hippocampus and superior temporal gyrus), as well as the extratemporal regions (such as the insula and prefrontal lobe); however, no significant intergroup differences were observed using volume-based fALFF. No significant correlation between fALFF and radiation dose to the ipsilateral temporal lobe was observed. Support vector machine (SVM) analysis revealed that surface-based fALFF in the bilateral superior temporal gyri and left insula exhibited impressive performance (accuracy = 80.49%) in identifying patients likely to develop RE. We conclude that surface-based fALFF may serve as a sensitive imaging biomarker in the prediction of RE.

Deep Learning for Classification of Bone Lesions on Routine MRI.

BACKGROUND: Radiologists have difficulty distinguishing benign from malignant bone lesions because these lesions may have similar imaging appearances. The purpose of this study was to develop a deep learning algorithm that can differentiate benign and malignant bone lesions using routine magnetic resonance imaging (MRI) and patient demographics. METHODS: 1,060 histologically confirmed bone lesions with T1- and T2-weighted pre-operative MRI were retrospectively identified and included, with lesions from 4 institutions used for model development and internal validation, and data from a fifth institution used for external validation. Image-based models were generated using the EfficientNet-B0 architecture and a logistic regression model was trained using patient age, sex, and lesion location. A voting ensemble was created as the final model. The performance of the model was compared to classification performance by radiology experts. FINDINGS: The cohort had a mean age of 30±23 years and was 58.3% male, with 582 benign lesions and 478 malignant. Compared to a contrived expert committee result, the ensemble deep learning model achieved (ensemble vs. experts): similar accuracy (0·76 vs. 0·73, p=0·7), sensitivity (0·79 vs. 0·81, p=1·0) and specificity (0·75 vs. 0·66, p=0·48), with a ROC AUC of 0·82. On external testing, the model achieved ROC AUC of 0·79. INTERPRETATION: Deep learning can be used to distinguish benign and malignant bone lesions on par with experts. These findings could aid in the development of computer-aided diagnostic tools to reduce unnecessary referrals to specialized centers from community clinics and limit unnecessary biopsies. FUNDING: This work was funded by a Radiological Society of North America Research Medical Student Grant (#RMS2013) and supported by the Amazon Web Services Diagnostic Development Initiative.

Prognostication of patients with COVID-19 using artificial intelligence based on chest x-rays and clinical data: a retrospective study.

BACKGROUND: Chest x-ray is a relatively accessible, inexpensive, fast imaging modality that might be valuable in the prognostication of patients with COVID-19. We aimed to develop and evaluate an artificial intelligence system using chest x-rays and clinical data to predict disease severity and progression in patients with COVID-19. METHODS: We did a retrospective study in multiple hospitals in the University of Pennsylvania Health System in Philadelphia, PA, USA, and Brown University affiliated hospitals in Providence, RI, USA. Patients who presented to a hospital in the University of Pennsylvania Health System via the emergency department, with a diagnosis of COVID-19 confirmed by RT-PCR and with an available chest x-ray from their initial presentation or admission, were retrospectively identified and randomly divided into training, validation, and test sets (7:1:2). Using the chest x-rays as input to an EfficientNet deep neural network and clinical data, models were trained to predict the binary outcome of disease severity (ie, critical or non-critical). The deep-learning features extracted from the model and clinical data were used to build time-to-event models to predict the risk of disease progression. The models were externally tested on patients who presented to an independent multicentre institution, Brown University affiliated hospitals, and compared with severity scores provided by radiologists. FINDINGS: 1834 patients who presented via the University of Pennsylvania Health System between March 9 and July 20, 2020, were identified and assigned to the model training (n=1285), validation (n=183), or testing (n=366) sets. 475 patients who presented via the Brown University affiliated hospitals between March 1 and July 18, 2020, were identified for external testing of the models. When chest x-rays were added to clinical data for severity prediction, area under the receiver operating characteristic curve (ROC-AUC) increased from 0·821 (95% CI 0·796-0·828) to 0·846 (0·815-0·852; p<0·0001) on internal testing and 0·731 (0·712-0·738) to 0·792 (0·780-0 ·803; p<0·0001) on external testing. When deep-learning features were added to clinical data for progression prediction, the concordance index (C-index) increased from 0·769 (0·755-0·786) to 0·805 (0·800-0·820; p<0·0001) on internal testing and 0·707 (0·695-0·729) to 0·752 (0·739-0·764; p<0·0001) on external testing. The image and clinical data combined model had significantly better prognostic performance than combined severity scores and clinical data on internal testing (C-index 0·805 vs 0·781; p=0·0002) and external testing (C-index 0·752 vs 0·715; p<0·0001). INTERPRETATION: In patients with COVID-19, artificial intelligence based on chest x-rays had better prognostic performance than clinical data or radiologist-derived severity scores. Using artificial intelligence, chest x-rays can augment clinical data in predicting the risk of progression to critical illness in patients with COVID-19. FUNDING: Brown University, Amazon Web Services Diagnostic Development Initiative, Radiological Society of North America, National Cancer Institute and National Institute of Biomedical Imaging and Bioengineering of the National Institutes of Health.

Pre-symptomatic local brain activity and functional connectivity alterations in nasopharyngeal carcinoma patients who developed radiation encephalopathy following radiotherapy.

Radiation encephalopathy (RE) is a common complication in patients with nasopharyngeal carcinoma (NPC) who have received radiotherapy (RT), and recent neuroimaging studies have shown brain alterations in Post-RT patients prior to RE. However, whether there are functional alterations between those Post-RT patients who are proved to have RE in follow-up and those who do not develop it remains largely unknown. Here, we used resting state functional MRI to explore regional homogeneity (ReHo) and functional connectivity (FC) alterations in Post-RT patients with (Post-RT RE proved; n = 18) or without (Post-RT non-RE; n = 22) RE at follow-up, also making comparisons with a Pre-RT group (n = 23). Compared with the Pre-RT group, patients in Post-RT non-RE and Post-RT RE proved groups showed concurrent increased and decreased ReHo values in different brain regions inside and/or outside the radiation field, with the alterations in ReHo tending to increase if RE occurred. Seed-based FC analysis showed that compared with the Post-RT non-RE group, patients in the Post-RT RE proved group had different changing patterns of FC between a region of interest (ROI) in the right temporal lobe and distant brain regions (mainly in the sensorimotor system and default mode network). Receiver operating characteristic (ROC) curve analysis showed that the altered ReHo value in the ROI had excellent diagnostic performance for differentiating NPC patients who developed RE in follow-up from those who did not, with an area under the curve (AUC) value of 0.94. These ReHo and FC findings may provide new insights into the early diagnosis of RE.

Prognostic Factors for Luminal B-like Breast Cancer.

This study aimed to examine the prognostic factors of luminal B-like breast cancer. Clinical data of 695 luminal B-like breast cancer patients who had been treated in our hospital during the period of past 4.5 years were collected and analyzed. Estrogen receptor (ER), progesterone receptor (PgR), antigen identified by monoclonal antibody Ki-67 (Ki67) were immunohistochemically detected. Different cutoffs of ER, PgR, and Ki67 were evaluated. Pearson χ2 test was performed to compare categorical parameters. Univariate and multivariate models were used to evaluate predictors of disease free survival (DFS). The results showed that patients who were younger, and had larger tumors, and more positive lymph nodes were more likely to receive neo-adjuvent chemotherapy (NAC). Patients with ER-positive tumors having <10% positive cells received more anthracycline- and taxane-based chemotherapy and less endocrine therapy than those with ER-positive tumors having ≥10% positive cells (P=0.004 and P=0.007, respectively); however, patients with ER-positive tumors having <10% positive cells experienced more recurrence (P<0.001). PgR expression levels were not associated with therapeutic schedule and DFS. Patients with tumor tissue Ki67 score ≥30% received more anthracycline- and taxane-based chemotherapy and had worse DFS than those with tumor tissue Ki67 score <30%. Univariate and multivariate analysis showed that clinical T stage, lymph nodes, ER, Ki67, and HER2 status were independent prognostic factors. In conclusion, ER-positive rate <10% and Ki67 score ≥30%, similar to higher clinical T stage, more metastatic lymph nodes, and HER2 positive status, may indicate a worse prognosis for luminal B-like breast cancer patients. Multi-center prospective trials with larger sample sizes are necessary for the continued perfection of our work.

A Predictive Scoring Model for Short-Term Local Recurrent Nasopharyngeal Carcinoma Based on Magnetic Resonance Imaging.

OBJECTIVE: To predict the early identification of recurrence based on magnetic resonance imaging (MRI) in nasopharyngeal cancer (NPC) patients. METHODS: The clinical and MRI data of 215 patients with local recurrent NPC were retrospectively reviewed. Logistic regression analysis was performed to distinguish the independent risk factors for the short-term (less than 24 months) local recurrence of NPC. The predictive score model was based on the regression coefficients of significant independent variables. RESULTS: Residual disease in the nasopharyngeal cavity (NC), masticator space invasion (MSI), skull base bone erosion (SBBE), and MRI-detected cranial nerve invasion (MDCNI) were all significant independent risk factors for the short-term recurrence of NPC (p < 0.05). The receiver operating characteristic curve showed that the total score had a maximal AUC (area under the curve) value of 0.897, with a cutoff point of 10.50. The sensitivity and specificity were 79.4% and 80.5%, respectively. CONCLUSION: Residual lesions in NC, MSI, SBBE, and MDCNI are independent risk factors in predicting the short-term recurrence of NPC. The authors' findings suggest that patients with a score of more than 10.50 points should be hypervigilant regarding the possibility of short-term recurrence.

Cortical Surface Area Rather Than Cortical Thickness Potentially Differentiates Radiation Encephalopathy at Early Stage in Patients With Nasopharyngeal Carcinoma.

Radiation encephalopathy (RE) is one of the most severe complications in nasopharyngeal carcinoma (NPC) patients after radiotherapy (RT). However, the morphological alteration of early RE is insufficiently investigated. We aimed to investigate the cortical thickness and surface area alterations in NPC patients with or without RE in the follow-up. A total of 168 NPC patients each underwent a single scan and analysis at various times either Pre-RT (n = 56) or Post-RT (n = 112). We further divided the Post-RT NPC patients into three groups based on the time of the analysis following RT (Post-RTwithin 6 months and Post-RT7-12 months) or whether RE signs were detected in the analysis (Post-RTRE proved in follow-up). We confined the vertex-wise analyses of the cortical thickness and surface area to the bilateral temporal lobes. Interestingly, we revealed a gradual increase in the cortical surface area of the temporal lobe with increasing time after RT within the Post-RTRE proved in follow-up group, consistent with the between-group findings, which showed a significant increase in cortical surface area in the Post-RTRE proved in follow-up group relative to the Pre-RT group and the Post-RTwithin 6 months group. By contrast, such a trend was not observed in the cortical thickness findings. We concluded that the cortical surface area, rather than cortical thickness, may serve as a potential biomarker for early diagnosis of RE.

NEK4 kinase regulates EMT to promote lung cancer metastasis.

Epithelial-to-mesenchymal transition (EMT) is a dynamic transitional state from the epithelial to mesenchymal phenotypes. Numerous studies have suggested that EMT and its intermediate states play important roles in tumor invasion and metastasis. To identify novel regulatory molecules of EMT, we screened a siRNA library targeting human 720 kinases in A549 lung adenocarcinoma cells harboring E-cadherin promoter-luciferase reporter vectors. NIMA-related kinase-4 (NEK4) was identified and characterized as a positive regulator of EMT in the screening. Suppression of NEK4 resulted in the inhibition of cell migration and invasion, accompanying with an increased expression of cell adhesion-related proteins such as E-cadherin and ZO1. Furthermore, NEK4 knockdown caused the decreased expression of the transcriptional factor Zeb1 and Smads proteins, which are known to play key roles in EMT regulation. Consistently, overexpression of NEK4 resulted in the decreased expression of E-cadherin and increased expression of Smad3. Using a mouse model with tail vein injection of NEK4 knockdown stable cell line, we found a lower rate of tumor formation and metastasis of the NEK4-knockdown cells in vivo. Thus, this study demonstrates NEK4 as a novel kinase involved in regulation of EMT and suggests that NEK4 may be further explored as a potential therapeutic target for lung cancer metastasis.

The imaging features of neurologic complications of left atrial myxomas.

BACKGROUND: Neurologic complications may be the first symptoms of atrial myxomas. Understanding the imaging features of neurologic complications of atrial myxomas can be helpful for the prompt diagnosis. OBJECTIVE: To identify neuroimaging features for patients with neurologic complications attributed to atrial myxoma. METHODS: We retrospectively reviewed the medical records of 103 patients with pathologically confirmed atrial myxoma at Xiangya Hospital from January 2009 to January 2014. The neuroimaging data for patients with neurologic complications were analyzed. RESULTS: Eight patients with atrial myxomas (7.77%) presented with neurologic manifestations, which constituted the initial symptoms for seven patients (87.5%). Neuroimaging showed five cases of cerebral infarctions and three cases of aneurysms. The main patterns of the infarctions were multiplicity (100.0%) and involvement of the middle cerebral artery territory (80.0%). The aneurysms were fusiform in shape, multiple in number (100.0%) and located in the distal middle cerebral artery (100.0%). More specifically, high-density in the vicinity of the aneurysms was observed on CT for two patients (66.7%), and homogenous enhancement surrounding the aneurysms was detected in the enhanced imaging for two patients (66.7%). CONCLUSION: Neurologic complications secondary to atrial myxoma consist of cerebral infarctions and aneurysms, which show certain characteristic features in neuroimaging. Echocardiography should be performed in patients with multiple cerebral infarctions, and multiple aneurysms, especially when aneurysms are distal in location. More importantly, greater attention should be paid to the imaging changes surrounding the aneurysms when myxomatous aneurysms are suspected and these are going to be the relevant features in our article.

DCE-MRI assessment of the effect of Epstein-Barr virus-encoded latent membrane protein-1 targeted DNAzyme on tumor vasculature in patients with nasopharyngeal carcinomas.

BACKGROUND: EBV-encoded latent membrane protein 1 (EBV-LMP1) is an important oncogenic protein for nasopharyngeal carcinoma (NPC) and has been shown to engage a plethora of signaling pathways. Correspondingly, an LMP1-targeted DNAzyme was found to inhibit the growth of NPC cells both in vivo and in vitro by suppressing cell proliferation and inducing apoptosis. However, it remains unknown whether an LMP1-targeted DNAzyme would affect the vasculature of NPC. Dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) has been applied in the clinical trials of anti-angiogenic drugs for more than ten years, and Ktrans has been recommended as a primary endpoint. Therefore, the objective of the current study was to use DCE-MRI to longitudinally study the effect of an EBV-LMP1-targeted DNAzyme on the vasculature of patients with NPC. METHODS: Twenty-four patients were randomly divided into two groups: a combined treatment group (radiotherapy + LMP1-targeted DNAzyme) and a radiotherapy alone group (radiotherapy + normal saline). DCE-MRI scans were conducted 1 ~ 2 days before radiotherapy (Pre-RT), during radiotherapy (RT 50 Gy), upon completion of radiotherapy (RT 70 Gy), and three months after radiotherapy (3 months post-RT). Parameters of vascular permeability and intra- and extravascular volumes were subsequently obtained (e.g., Ktrans, kep, ve) using nordicICE software. RESULTS: Both Ktrans and kep values for NPC tumor tissues decreased for both groups after treatment. Moreover, a statistically significant difference in Ktrans values at the pre-therapy and post-therapy timepoints emerged earlier for the combined treatment group (RT 50 Gy, P =0.045) compared to the radiotherapy alone group (3 months post-RT, P = 0.032). For the kep values, the downward trend observed for both the combined treatment group and the radiotherapy alone group were similar. In contrast, ve values for all of the tumor tissues increased following therapy. CONCLUSIONS: The EBV-LMP1-targeted DNAzyme that was tested was found to accelerate the decline of Ktrans values for patients with NPC. Correspondingly, the LMP1-targeted DNAzyme treatments were found to affect the angiogenesis and microvascular permeability of NPC. TRIAL REGISTRATION: ClinicalTrials.gov: NCT01449942. Registered 6 October 2011.

NK/T cell lymphoma involving mediastinum: report of a case and review of literature.

Extranodal natural killer (NK)/T-cell lymphoma is a very aggressive malignant neoplasia with a poor prognosis. Herein we reported a case of NK/T cell lymphoma involving mediastinum. It was a 28-year-old Chinese male patient. The tumor cells were medium-sized, had irregularly folded nuclei, and inconspicuous or small nucleoli with coagulative necrosis. The tumor cells were positive for CD3ε, TIA-1, but negative for CD56. In situ hybridization revealed that tumor cells also expressed Epstein-Barr virus encoded RNA. To our knowledge, this is the first case of NK/T cell lymphoma involving mediastinum.

[Correlation between dynamic contrast-enhanced MRI and expression of microvascular density in the glioma].

OBJECTIVE: To explore the correlation between microvascular density (MVD) and dynamic contrast-enhanced MRI in the glioma. METHODS: We examined 35 patients with histologically verified glioma. Gadolinium-enhanced dynamic TurboFLASH imaging was performed preoperatively in all patients followed by conventional MRI. The steepest slope (SSmax) of curve and corresponding Tm1 in "first-pass" phase were obtained by analyzing time-signal curve. All specimens were immunostained with anti-human Factor VIII relative antigen monoclonal antibody postoperatively by streptavidin-peroxidase method to obtain the MVD. The correlation between SSmax, Tm1, and MVD was analyzed. RESULTS: SSmax was positively correlated with MVD (r = 0.640, P < 0.01). Tml was negatively correlated with MVD (r = -0.671, P < 0.01). CONCLUSION: The MVD correlates obviously with SSmax and Tml in the glioma. Analyzing the time-signal curve of dynamic contrast-enhanced MRI is helpful to predict the angiogenesis in the glioma.