Accelerated MRI reconstructions via variational network and feature domain learning.

We introduce three architecture modifications to enhance the performance of the end-to-end (E2E) variational network (VarNet) for undersampled MRI reconstructions. We first implemented the Feature VarNet, which propagates information throughout the cascades of the network in an N-channel feature-space instead of a 2-channel feature-space. Then, we add an attention layer that utilizes the spatial locations of Cartesian undersampling artifacts to further improve performance. Lastly, we combined the Feature and E2E VarNets into the Feature-Image (FI) VarNet, to facilitate cross-domain learning and boost accuracy. Reconstructions were evaluated on the fastMRI dataset using standard metrics and clinical scoring by three neuroradiologists. Feature and FI VarNets outperformed the E2E VarNet for 4 × , 5 × and 8 × Cartesian undersampling in all studied metrics. FI VarNet secured second place in the public fastMRI leaderboard for 4 × Cartesian undersampling, outperforming all open-source models in the leaderboard. Radiologists rated FI VarNet brain reconstructions with higher quality and sharpness than the E2E VarNet reconstructions. FI VarNet excelled in preserving anatomical details, including blood vessels, whereas E2E VarNet discarded or blurred them in some cases. The proposed FI VarNet enhances the reconstruction quality of undersampled MRI and could enable clinically acceptable reconstructions at higher acceleration factors than currently possible.

Inter-reader agreement of pancreatic adenocarcinoma resectability assessment with photon counting versus energy integrating detector CT.

PURPOSE: To compare the inter-reader agreement of pancreatic adenocarcinoma resectability assessment at pancreatic protocol photon-counting CT (PCCT) with conventional energy-integrating detector CT (EID-CT). METHODS: A retrospective single institution database search identified all contrast-enhanced pancreatic mass protocol abdominal CT performed at an outpatient facility with both a PCCT and EID-CT from 4/11/2022 to 10/30/2022. Patients without pancreatic adenocarcinoma were excluded. Four fellowship-trained abdominal radiologists, blinded to CT type, independently assessed vascular tumor involvement (uninvolved, abuts ≤ 180°, encases > 180°; celiac, superior mesenteric artery (SMA), common hepatic artery (CHA), superior mesenteric vein (SMV), main portal vein), the presence/absence of metastases, overall tumor resectability (resectable, borderline resectable, locally advanced, metastatic), and diagnostic confidence. Fleiss's kappa was used to calculate inter-reader agreement. CTDIvol was recorded. Radiation dose metrics were compared with a two-sample t-test. A p < .05 indicated statistical significance. RESULTS: 145 patients (71 men, mean[SD] age: 66[9] years) were included. There was substantial inter-reader agreement, for celiac artery, SMA, and SMV involvement at PCCT (kappa = 0.61-0.69) versus moderate agreement at EID-CT (kappa = 0.56-0.59). CHA had substantial inter-reader agreement at both PCCT (kappa = 0.67) and EIDCT (kappa = 0.70). For metastasis identification, radiologists had substantial inter-reader agreement at PCCT (kappa = 0.78) versus moderate agreement at EID-CT (kappa = 0.56). CTDIvol for PCCT and EID-CT were 16.9[7.4]mGy and 29.8[26.6]mGy, respectively (p < .001). CONCLUSION: There was substantial inter-reader agreement for involvement of 4/5 major peripancreatic vessels (celiac artery, SMA, CHA, and SMV) at PCCT compared with 2/5 for EID-CT. PCCT also afforded substantial inter-reader agreement for metastasis detection versus moderate agreement at EID-CT with statistically significant radiation dose reduction.

Pancreatic cyst prevalence and detection with photon counting CT compared with conventional energy integrating detector CT.

PURPOSE: To calculate the prevalence of pancreatic cysts on photon counting CT (PCCT) and compare with that of 128-slice conventional energy-integrating detector CT (EIDCT). METHOD: A retrospective single institution database search identified all contrast-enhanced abdominal CT examinations performed at an outpatient facility that has both a PCCT and EIDCT between 4/11/2022 and 7/26/2022. The presence and size of pancreatic cysts were recorded. In patients with PCCT reported pancreatic cysts, prior CT imaging (EIDCT) was reviewed for reported pancreatic cysts. Fisher's exact test was used to compare the pancreatic cyst detection rate for PCCT and EIDCT. Wilcoxon rank sum test was used to compare cyst size and patient age. A p <.05 indicated statistical significance. RESULTS: 2494 patients were included. Our pancreatic cyst detection rate was 4.9 % (49/1009) with PCCT and 3.0 % (44/1485) for EIDCT (p =.017). For CT angiograms, pancreatic cysts were detected in 6.6 % (21/319) with PCCT and 0.0 % (0/141) with EIDCT (p <.001). Pancreatic cyst detection rate was not statistically different for portal venous, enterography, renal mass, pancreas, 3-phase liver, or venogram protocols (all p >.05). Mean[SD] pancreatic cyst size was 13.7[9.7]mm for PCCT and 15.3[14.7] for EIDCT (p =.95). 55.1 % (27/49) of PCCT and 61.4 % (27/44) of EIDCT that described pancreatic cysts had prior contrast-enhanced EIDCTs. Of these, 40.7 % (11/27) of PCCT and 14.8 % (4/27) of EIDCT described pancreatic cysts were not previously reported (p =.027). CONCLUSIONS: Photon-counting CT afforded greater pancreatic cyst detection than conventional energy-integrating detector CT, particularly with CT angiograms.

Improving Information Extraction from Pathology Reports using Named Entity Recognition.

Pathology reports are considered the gold standard in medical research due to their comprehensive and accurate diagnostic information. Natural language processing (NLP) techniques have been developed to automate information extraction from pathology reports. However, existing studies suffer from two significant limitations. First, they typically frame their tasks as report classification, which restricts the granularity of extracted information. Second, they often fail to generalize to unseen reports due to variations in language, negation, and human error. To overcome these challenges, we propose a BERT (bidirectional encoder representations from transformers) named entity recognition (NER) system to extract key diagnostic elements from pathology reports. We also introduce four data augmentation methods to improve the robustness of our model. Trained and evaluated on 1438 annotated breast pathology reports, acquired from a large medical center in the United States, our BERT model trained with data augmentation achieves an entity F1-score of 0.916 on an internal test set, surpassing the BERT baseline (0.843). We further assessed the model's generalizability using an external validation dataset from the United Arab Emirates, where our model maintained satisfactory performance (F1-score 0.860). Our findings demonstrate that our NER systems can effectively extract fine-grained information from widely diverse medical reports, offering the potential for large-scale information extraction in a wide range of medical and AI research. We publish our code at https://github.com/nyukat/pathology_extraction.

Detection of Incidental Nonosseous Thoracic Pathology on State-of-the-Art Ultralow-Dose Protocol Computed Tomography in Pediatric Patients With Pectus Excavatum.

OBJECTIVE: The aim of the study was to compare a pediatric ultralow-dose pectus excavatum computed tomography (CT) protocol versus standard-dose pediatric thoracic CT in terms of radiation dose, subjective and objective image quality, and its ability to detect incidental nonosseous thoracic pathology compared with imaging and clinical reference. METHODS: A single institution radiology database identified a total of 104 ultralow-dose pediatric thoracic CT cases with an equal number of age-matched standard-dose chest CT cases also selected for retrospective analysis. Objective image quality (contrast-to-noise and signal-to-noise ratios) and radiation dose were assessed. Qualitative Likert scorings of the bone, lung, and soft tissues were performed by 2 expert radiologists. Electronic health records of the ultralow-dose cohort were reviewed for at least 1 year to evaluate for potentially missed thoracic pathology and symptoms. Variables were compared using parametric and nonparametric tests in R software 4.0.5. RESULTS: The ultralow-dose protocol group had statistically significant reductions (P < 0.001) in the volume CT dose index (0.31 ± 0.19 vs 2.20 ± 1.64 mGy), effective radiation dose (0.14 ± 0.08 vs 1.07 ± 0.86 mSv), and size-specific dose estimates (0.50 ± 0.30 vs 3.43 ± 2.56 mGy) compared with the standard protocol, yielding an 86.51% and 85.32% reduction, respectively. The signal-to-noise ratio (20.49 ± 6.19 vs 36.48 ± 10.20), contrast-to-noise (21.65 ± 6.57 vs 38.47 ± 10.59), and subjective measures of image quality (lung parenchyma [3.07 ± 0.92 vs 4.42 ± 0.47], bony structures [3.30 ± 0.86 vs 4.52 ± 0.51], and surrounding soft tissues [2.57 ± 0.63 vs 3.89 ± 0.65]) were also significantly lower in the ultralow-dose protocol (P < 0.001). No differences were seen in the number and size of pulmonary nodules between groups. Clinical and imaging follow of all 104 patients undergoing ultralow-dose CT demonstrated no evidence of missed thoracic pathology causing symptoms. CONCLUSIONS: Ultralow-dose thoracic CT is an acceptable modality for imaging pediatric patients with pectus excavatum and other conditions primarily causing osseous pathology, with effective radiation dose comparable to plain radiographs and a moderate increase in image noise that did not significantly reduce its ability to detect incidental nonosseous thoracic pathology.

Abdominal and testicular pain: An atypical presentation of COVID-19.

The outbreak of a novel coronavirus disease (COVID-19) has been of concern to health care workers (HCW's) in the emergency department (ED) due to potential exposure and transmission. This case report describes a man who was referred to the ED for abdominal and testicular pain who was subsequently found to test positive for COVID-19. Due to the lack of respiratory symptoms, proper protective equipment (PPE) was not donned, and it led to several patients and health care workers being exposed. Given recent new descriptions of patients who present atypically, full PPE for all patients may be considered as community spread increases.

A dominant-negative effect drives selection of TP53 missense mutations in myeloid malignancies.

TP53, which encodes the tumor suppressor p53, is the most frequently mutated gene in human cancer. The selective pressures shaping its mutational spectrum, dominated by missense mutations, are enigmatic, and neomorphic gain-of-function (GOF) activities have been implicated. We used CRISPR-Cas9 to generate isogenic human leukemia cell lines of the most common TP53 missense mutations. Functional, DNA-binding, and transcriptional analyses revealed loss of function but no GOF effects. Comprehensive mutational scanning of p53 single-amino acid variants demonstrated that missense variants in the DNA-binding domain exert a dominant-negative effect (DNE). In mice, the DNE of p53 missense variants confers a selective advantage to hematopoietic cells on DNA damage. Analysis of clinical outcomes in patients with acute myeloid leukemia showed no evidence of GOF for TP53 missense mutations. Thus, a DNE is the primary unit of selection for TP53 missense mutations in myeloid malignancies.

Detecting hierarchical genome folding with network modularity.

Mammalian genomes are folded in a hierarchy of compartments, topologically associating domains (TADs), subTADs and looping interactions. Here, we describe 3DNetMod, a graph theory-based method for sensitive and accurate detection of chromatin domains across length scales in Hi-C data. We identify nested, partially overlapping TADs and subTADs genome wide by optimizing network modularity and varying a single resolution parameter. 3DNetMod can be applied broadly to understand genome reconfiguration in development and disease.

Local Genome Topology Can Exhibit an Incompletely Rewired 3D-Folding State during Somatic Cell Reprogramming.

Pluripotent genomes are folded in a topological hierarchy that reorganizes during differentiation. The extent to which chromatin architecture is reconfigured during somatic cell reprogramming is poorly understood. Here we integrate fine-resolution architecture maps with epigenetic marks and gene expression in embryonic stem cells (ESCs), neural progenitor cells (NPCs), and NPC-derived induced pluripotent stem cells (iPSCs). We find that most pluripotency genes reconnect to target enhancers during reprogramming. Unexpectedly, some NPC interactions around pluripotency genes persist in our iPSC clone. Pluripotency genes engaged in both "fully-reprogrammed" and "persistent-NPC" interactions exhibit over/undershooting of target expression levels in iPSCs. Additionally, we identify a subset of "poorly reprogrammed" interactions that do not reconnect in iPSCs and display only partially recovered, ESC-specific CTCF occupancy. 2i/LIF can abrogate persistent-NPC interactions, recover poorly reprogrammed interactions, reinstate CTCF occupancy, and restore expression levels. Our results demonstrate that iPSC genomes can exhibit imperfectly rewired 3D-folding linked to inaccurately reprogrammed gene expression.