Less Likely Brainstorming: Using Language Models to Generate Alternative Hypotheses.

A human decision-maker benefits the most from an AI assistant that corrects for their biases. For problems such as generating interpretation of a radiology report given findings, a system predicting only highly likely outcomes may be less useful, where such outcomes are already obvious to the user. To alleviate biases in human decision-making, it is worth considering a broad differential diagnosis, going beyond the most likely options. We introduce a new task, "less likely brainstorming," that asks a model to generate outputs that humans think are relevant but less likely to happen. We explore the task in two settings: a brain MRI interpretation generation setting and an everyday commonsense reasoning setting. We found that a baseline approach of training with less likely hypotheses as targets generates outputs that humans evaluate as either likely or irrelevant nearly half of the time; standard MLE training is not effective. To tackle this problem, we propose a controlled text generation method that uses a novel contrastive learning strategy to encourage models to differentiate between generating likely and less likely outputs according to humans. We compare our method with several state-of-the-art controlled text generation models via automatic and human evaluations and show that our models' capability of generating less likely outputs is improved.

Evaluating large language models on medical evidence summarization.

Recent advances in large language models (LLMs) have demonstrated remarkable successes in zero- and few-shot performance on various downstream tasks, paving the way for applications in high-stakes domains. In this study, we systematically examine the capabilities and limitations of LLMs, specifically GPT-3.5 and ChatGPT, in performing zero-shot medical evidence summarization across six clinical domains. We conduct both automatic and human evaluations, covering several dimensions of summary quality. Our study demonstrates that automatic metrics often do not strongly correlate with the quality of summaries. Furthermore, informed by our human evaluations, we define a terminology of error types for medical evidence summarization. Our findings reveal that LLMs could be susceptible to generating factually inconsistent summaries and making overly convincing or uncertain statements, leading to potential harm due to misinformation. Moreover, we find that models struggle to identify the salient information and are more error-prone when summarizing over longer textual contexts.

The effect of aminolevulinic acid hydrochloride topical powder photodynamic therapy versus nocardia rubra cell wall skeleton and observers group in the treatment of cervical low-grade squamous intraepithelial lesions with human papillomavirus infection: A non-randomized study, controlled pilot study.

There are insufficient studies comparing the efficacy of aminolevulinic acid hydrochloride topical powder (ALA) photodynamic therapy (PDT) against Nocardia rubra cell wall skeleton (Nr-CWS) therapy in the treatment of cervical low-grade squamous intraepithelial lesion (LSIL) with human papillomavirus (HPV), especially for long-term efficacy. Patients with cervical LSIL and HPV infection were divided into 3 treatment groups based on their own choice. All patients had a follow-up test including HPV testing, cytology and colposcopy at 4 to 6 months and 12 months after the treatment. Among142 patients, patients received 51 ALA PDT and 41 patients received Nr-CWS. Another 50 patients who refused treatment were included in the Observers group. Four to six months or 12 months after treatment, there was significant difference between 3 groups in the clearance rate of HR-HPV infection and the complete remission (CR) rates of cervical LSIL; the CR rates of cervical LSIL in the ALA PDT group was significantly higher than the Nr-CWS group; but there was no significant difference between 2 groups in the clearance rate of HPV infection. The CR rates of cervical LSIL and the clearance rate of HPV infection in the ALA PDT group was significantly higher than the Observers group; the CR rates of cervical LSIL and the clearance rate of HPV infection in the Nr-CWS group was significantly higher than the Observers group; there was no significant difference in the recurrence rates in ALA PDT and Nr-CWS group after 12 months. Both of ALA PDT and Nr-CWS group had lower recurrence rate than the Observers group. The effect of ALA PDT is similar to Nr-CWS in the clearance rate of HR-HPV infection. Compared to the Nr-CWS group, the CR rates of cervical LSIL were considerably greater in the ALA PDT group. The effect of ALA PDT in the clearance rate of HPV infection and CR rates of cervical LSIL was significantly higher than the follow-up group; Both of ALA PDT and Nr-CWS group had lower recurrence rate than the Observers group. For cervical LSIL with HPV infection, ALA PDT is a very successful therapeutic method that is noninvasive.

Evaluating Large Language Models on Medical Evidence Summarization.

Recent advances in large language models (LLMs) have demonstrated remarkable successes in zero- and few-shot performance on various downstream tasks, paving the way for applications in high-stakes domains. In this study, we systematically examine the capabilities and limitations of LLMs, specifically GPT-3.5 and ChatGPT, in performing zero-shot medical evidence summarization across six clinical domains. We conduct both automatic and human evaluations, covering several dimensions of summary quality. Our study has demonstrated that automatic metrics often do not strongly correlate with the quality of summaries. Furthermore, informed by our human evaluations, we define a terminology of error types for medical evidence summarization. Our findings reveal that LLMs could be susceptible to generating factually inconsistent summaries and making overly convincing or uncertain statements, leading to potential harm due to misinformation. Moreover, we find that models struggle to identify the salient information and are more error-prone when summarizing over longer textual contexts.

Effects of gross total resection and subtotal resection on survival outcomes of glioma patients: a meta-analysis.

This study was to conduct a meta-analysis to explore the impact of gross total resection (GTR) and subtotal resection (STR) on survival outcomes in glioma patients. Relevant studies were searched in multiple databases from the available date of inception through 30 December 2021. The weighted mean differences (WMDs), relative risks (RRs), or hazard ratios (HRs) with 95% confidence intervals (CIs) were used to access the effect of GTR versus STR treatments on the outcomes. The histology (low-grade or high-grade) and study population (children and adults) were used for subgroup analysis. Sensitivity analysis was performed for all outcomes. Begg's test and trim-and-fill method were used for publication bias. Totally 100 studies enrolling 62,129 patients were selected in this meta-analysis. The summary results showed that GTR was superior in improving 1-, 2-, 3-, 5-, 10-, 15-year overall survival (OS), OS time, 1-, 3-, 5-year progression-free survival (PFS), recurrence, local control and seizure control among glioma patients. In addition, high-grade patients who underwent GTR had improvements in 1-, 2- and 3-year OS, OS time, and 1-year PFS, while low-grade patients receiving GTR had improvements in 2-, 5- and 15-year OS, recurrence, seizure control, and tumor progression compared with those receiving STR. GTR was likely to be more effective on survival outcomes than STR among patients with gliomas.

Effects of dietary restriction on genome stability are sex and feeding regimen dependent.

Preserving genome stability is essential to prevent aging and cancer. Dietary restriction (DR) is the most reproducible non-pharmacological way to improve health and extend lifespan in various species. Whether DR helps to preserve genome stability and whether this effect is altered by experimental variables remain unclear. Moreover, DR research relies heavily on experimental animals, making the development of reliable in vitro mimetics of great interest. Therefore, we tested the effects of sex and feeding regimen (time-restricted eating, alternate day fasting and calorie restriction) on genome stability in CF-1 mice and whether these effects can be recapitulated by cell culture paradigms. Here, we show that calorie restriction significantly decreases the spontaneous micronuclei (MN), a biomarker of genome instability, in bone marrow cells of females instead of males. Alternate day fasting significantly decreases cisplatin-induced MN in females instead of males. Unexpectedly, daily time-restricted eating significantly exacerbates cisplatin-induced MN in males but not in females. Additionally, we design several culture paradigms that are able to faithfully recapitulate the key effects of these DR regimens on genome stability. In particular, 30% reduction of serum, a mimetic of calorie restriction, exhibits a strong ability to decrease spontaneous and cisplatin-induced MN in immortalized human umbilical vein endothelial cells. We conclude that the effects of different DR regimens on genome stability are not universal and females from each diet regimen sustain a more stable genome than males. Our results provide novel insight into the understanding of how DR influences genome stability in a sex and regimen dependent way, and suggest that our in vitro DR mimetics could be adopted to study the underlying molecular mechanisms.

Comparative effect of stenting plus medical therapy vs medical therapy alone on the risk of stroke and death in patients with symptomatic intracranial stenosis: a systematic review and meta-analysis.

BACKGROUND: Recently, several randomized controlled trials (RCTs) of stenting plus medical therapy versus medical therapy alone have been successfully conducted for the treatment of patients with symptomatic intracranial stenosis. This study aimed to evaluate differences between these two therapies in the risk of stroke and death. METHODS: MEDLINE, EMBASE, the Cochrane Library, and ClinicalTrials.gov were systematically searched to identify relevant studies published before August 24, 2022. Review Manager 5.3 software was used to assess the data. The risk ratio (RR) was analysed and calculated with a random effect model or a fixed effects model. RESULTS: We pooled 921 participants from three RCTs. Compared to the medical therapy alone group, the stenting plus medical therapy group had a higher risk of 30-day death or stroke (RR = 2.69 [1.64-4.41], P < 0.0001, I2 = 0%). When the follow-up period exceeded 1 year (≥ 1 year), there was no significant difference in the risk of stroke or death between these two groups. The subgroup analysis showed that if the time from stroke onset to implantation was extended, additional stenting would have no effect on the risk of stroke or death, whether within 30 days or within 1 year (P = 0.16 and 0.78). CONCLUSION: Medical therapy alone has a lower risk of stroke and death in the short term than stenting plus medical therapy, while no difference exists in the long term. More studies are still needed to further explore the precision strategy of stent implantation for symptomatic intracranial stenosis patients.

A glycosylation-related gene signature predicts prognosis, immune microenvironment infiltration, and drug sensitivity in glioma.

Glioma represents the most common primary cancer of the central nervous system in adults. Glycosylation is a prevalent post-translational modification that occurs in eukaryotic cells, leading to a wide array of modifications on proteins. We obtained the clinical information, bulk RNA-seq data, and single-cell RNA sequencing (scRNA-seq) from The Cancer Genome Atlas (TCGA), Chinese Glioma Genome Atlas (CGGA), Gene Expression Omnibus (GEO), and Repository of Molecular Brain Neoplasia Data (Rembrandt) databases. RNA sequencing data for normal brain tissues were accessed from the Genotype-Tissue Expression (GTEx) database. Then, the glycosylation genes that were differentially expressed were identified and further subjected to variable selection using a least absolute shrinkage and selection operator (LASSO)-regularized Cox model. We further conducted enrichment analysis, qPCR, nomogram, and single-cell transcriptome to detect the glycosylation signature. Drug sensitivity analysis was also conducted. A five-gene glycosylation signature (CHPF2, PYGL, GALNT13, EXT2, and COLGALT2) classified patients into low- or high-risk groups. Survival analysis, qPCR, ROC curves, and stratified analysis revealed worse outcomes in the high-risk group. Furthermore, GSEA and immune infiltration analysis indicated that the glycosylation signature has the potential to predict the immune response in glioma. In addition, four drugs (crizotinib, lapatinib, nilotinib, and topotecan) showed different responses between the two risk groups. Glioma cells had been classified into seven lines based on single-cell expression profiles. The five-gene glycosylation signature can accurately predict the prognosis of glioma and may offer additional guidance for immunotherapy.

EchoGen: A New Benchmark Study on Generating Conclusions from Echocardiogram Notes.

Generating a summary from findings has been recently explored (Zhang et al., 2018, 2020) in note types such as radiology reports that typically have short length. In this work, we focus on echocardiogram notes that is longer and more complex compared to previous note types. We formally define the task of echocardiography conclusion generation (EchoGen) as generating a conclusion given the findings section, with emphasis on key cardiac findings. To promote the development of EchoGen methods, we present a new benchmark, which consists of two datasets collected from two hospitals. We further compare both standard and state-of-the-art methods on this new benchmark, with an emphasis on factual consistency. To accomplish this, we develop a tool to automatically extract concept-attribute tuples from the text. We then propose an evaluation metric, FactComp, to compare concept-attribute tuples between the human reference and generated conclusions. Both automatic and human evaluations show that there is still a significant gap between human-written and machine-generated conclusions on echo reports in terms of factuality and overall quality.

Trustworthy assertion classification through prompting.

Accurate identification of the presence, absence or possibility of relevant entities in clinical notes is important for healthcare professionals to quickly understand crucial clinical information. This introduces the task of assertion classification - to correctly identify the assertion status of an entity in the unstructured clinical notes. Recent rule-based and machine-learning approaches suffer from labor-intensive pattern engineering and severe class bias toward majority classes. To solve this problem, in this study, we propose a prompt-based learning approach, which treats the assertion classification task as a masked language auto-completion problem. We evaluated the model on six datasets. Our prompt-based method achieved a micro-averaged F-1 of 0.954 on the i2b2 2010 assertion dataset, with ∼1.8% improvements over previous works. In particular, our model showed excellence in detecting classes with few instances (few-shot). Evaluations on five external datasets showcase the outstanding generalizability of the prompt-based method to unseen data. To examine the rationality of our model, we further introduced two rationale faithfulness metrics: comprehensiveness and sufficiency. The results reveal that compared to the "pre-train, fine-tune" procedure, our prompt-based model has a stronger capability of identifying the comprehensive (∼63.93%) and sufficient (∼11.75%) linguistic features from free text. We further evaluated the model-agnostic explanations using LIME. The results imply a better rationale agreement between our model and human beings (∼71.93% in average F-1), which demonstrates the superior trustworthiness of our model.

Prior Knowledge Enhances Radiology Report Generation.

Radiology report generation aims to produce computer-aided diagnoses to alleviate the workload of radiologists and has drawn increasing attention recently. However, previous deep learning methods tend to neglect the mutual influences between medical findings, which can be the bottleneck that limits the quality of generated reports. In this work, we propose to mine and represent the associations among medical findings in an informative knowledge graph and incorporate this prior knowledge with radiology report generation to help improve the quality of generated reports. Experiment results demonstrate the superior performance of our proposed method on the IU X-ray dataset with a ROUGE-L of 0.384±0.007 and CIDEr of 0.340±0.011. Compared with previous works, our model achieves an average of 1.6% improvement (2.0% and 1.5% improvements in CIDEr and ROUGE-L, respectively). The experiments suggest that prior knowledge can bring performance gains to accurate radiology report generation. We will make the code publicly available at https://github.com/bionlplab/report_generation_amia2022.

Measurement of a helium tune-out frequency: an independent test of quantum electrodynamics.

Despite quantum electrodynamics (QED) being one of the most stringently tested theories underpinning modern physics, recent precision atomic spectroscopy measurements have uncovered several small discrepancies between experiment and theory. One particularly powerful experimental observable that tests QED independently of traditional energy level measurements is the "tune-out" frequency, where the dynamic polarizability vanishes and the atom does not interact with applied laser light. In this work, we measure the tune-out frequency for the 23S1 state of helium between transitions to the 23P and 33P manifolds and compare it with new theoretical QED calculations. The experimentally determined value of 725,736,700(260) megahertz differs from theory [725,736,252(9) megahertz] by 1.7 times the measurement uncertainty and resolves both the QED contributions and retardation corrections.

Using Radiomics as Prior Knowledge for Thorax Disease Classification and Localization in Chest X-rays.

Chest X-ray becomes one of the most common medical diagnoses due to its noninvasiveness. The number of chest X-ray images has skyrocketed, but reading chest X-rays still have been manually performed by radiologists, which creates huge burnouts and delays. Traditionally, radiomics, as a subfield of radiology that can extract a large number of quantitative features from medical images, demonstrates its potential to facilitate medical imaging diagnosis before the deep learning era. In this paper, we develop an end-to-end framework, ChexRadiNet, that can utilize the radiomics features to improve the abnormality classification performance. Specifically, ChexRadiNet first applies a light-weight but efficient triplet-attention mechanism to classify the chest X-rays and highlight the abnormal regions. Then it uses the generated class activation map to extract radiomic features, which further guides our model to learn more robust image features. After a number of iterations and with the help of radiomic features, our framework can converge to more accurate image regions. We evaluate the ChexRadiNet framework using three public datasets: NIH ChestX-ray, CheXpert, and MIMIC-CXR. We find that ChexRadiNet outperforms the state-of-the-art on both disease detection (0.843 in AUC) and localization (0.679 in T(IoU) = 0.1). We make the code publicly available at https://github. com/bionlplab/lung_disease_detection_amia2021, with the hope that this method can facilitate the development of automatic systems with a higher-level understanding of the radiological world.

RadBERT-CL: Factually-Aware Contrastive Learning For Radiology Report Classification.

Radiology reports are unstructured and contain the imaging findings and corresponding diagnoses transcribed by radiologists which include clinical facts and negated and/or uncertain statements. Extracting pathologic findings and diagnoses from radiology reports is important for quality control, population health, and monitoring of disease progress. Existing works, primarily rely either on rule-based systems or transformer-based pre-trained model fine-tuning, but could not take the factual and uncertain information into consideration, and therefore generate false positive outputs. In this work, we introduce three sedulous augmentation techniques which retain factual and critical information while generating augmentations for contrastive learning. We introduce RadBERT-CL, which fuses these information into BlueBert via a self-supervised contrastive loss. Our experiments on MIMIC-CXR show superior performance of RadBERT-CL on fine-tuning for multi-class, multi-label report classification. We illustrate that when few labeled data are available, RadBERT-CL outperforms conventional SOTA transformers (BERT/BlueBert) by significantly larger margins (6-11%). We also show that the representations learned by RadBERT-CL can capture critical medical information in the latent space.

AoStuA, an APSES transcription factor, regulates the conidiation, trap formation, stress resistance and pathogenicity of the nematode-trapping fungus Arthrobotrys oligospora.

The APSES protein family comprises a conserved class of fungus-specific transcriptional regulators. Some members have been identified in partial ascomycetes. In this study, the APSES protein StuA (AoStuA) of the nematode-trapping fungus Arthrobotrys oligospora was characterized. Compared with the wild-type (WT) strain, three ΔAoStuA mutants grew relatively slowly, displayed a 96% reduction in sporulation capacity and a delay in conidial germination. The reduced sporulation capacity correlated with transcriptional repression of several sporulation-related genes. The mutants were also more sensitive to chemical stressors than the WT strain. Importantly, the mutants were unable to produce mycelial traps for nematode predation. Moreover, peroxisomes and Woronin bodies were abundant in the WT cells but hardly found in the cells of those mutants. The lack of such organelles correlated with transcriptional repression of some genes involved in the biogenesis of peroxisomes and Woronin bodies. The transcript levels of several genes involved in the cAMP/PKA signalling pathway were also significantly reduced in the mutants versus the WT strain, implicating a regulatory role of AoStuA in the transcription of genes involved in the cAMP/PKA signalling pathway that regulates an array of cellular processes and events. In particular, AoStuA is indispensable for A. oligospora trap formation and virulence.

Age-related network topological difference based on the sleep ECG signal.

OBJECTIVE: Age has been shown to be a crucial factor for the EEG and fMRI small-world networks during sleep. However, the characteristics of the age-related network based on the sleep ECG signal and how the network changes during different sleep stages are poorly understood. This study focuses on exploring the age-related scale-free and small-world network properties of the ECG signal from male subjects during distinct sleep stages, including the wakeful (W), light sleep (LS), deep sleep (DS) and rapid eye movement (REM) stages. APPROACH: The subjects are divided into two age groups: a younger (age ⩽ 40, n = 11) group and an older group (age > 40, n = 25). MAIN RESULTS: For the scale-free network analysis, our results reveal a distinctive pattern of the scale free network topologies between the two age groups, including the mean degree ([Formula: see text]), the clustering coefficient ([Formula: see text]), and the path length ([Formula: see text]) features, such as the slope distribution of [Formula: see text] in the younger group increased from 1.99 during W to above 2.05 during DS. In addition, the results indicate that the small-world properties can be found across all sleep stages in both age groups. However, the small-world index in the LS and REM stages significantly decreased with age (p = 0.0006 and p = 0.05, respectively). SIGNIFICANCE: The comparison analysis result indicates that the network topology variations in the sleep ECG signals are prone to show age-relevant differences that could be used for sleep stage classification and sleep disorder diagnosis.

Regulation of nonylphenol-induced reproductive toxicity in mouse spermatogonia cells by miR-361-3p.

Nonylphenol (NP) is an environmental chemical that affects apoptosis and male infertility. In our study, we found that a high concentration of NP could down-regulate the expression of microRNA-361-3p (miR-361-3p) in the murine GC-1 spermatogonia cell line and in vivo in murine spermatogonia. Additionally, one direct target of this miR, the 3' untranslated region of Killin (Klln) mRNA, was identified. Klln encodes a transcription factor that directly regulates the expression of Tp73 (transcriptionally active p73), whose encoded protein can up-regulate the expression of Puma (p53 upregulated modulator of apoptosis). Thus, our investigation revealed that the expression of Klln, Tp73, and Puma increased upon NP-dependent down-regulation of miR-361-3p, which eventually leads to apoptosis of spermatogonia.

Random mutagenesis analysis and identification of a novel C2H2-type transcription factor from the nematode-trapping fungus Arthrobotrys oligospora.

Arthrobotrys oligospora is a typical nematode-trapping fungus. In this study, 37 transformants of A. oligospora were obtained by REMI (restriction enzyme mediated integration) method and phenotypic properties of nine transformants were analyzed. The nine transformants showed differences in growth, conidiation, trap formation, stress tolerance, and/or pathogenicity among each other and with those of the parental wild-type strain (WT). The insertional sites of the hph cassette were identified in transformants X5 and X13. In X5, the cassette was inserted in the non-coding region between AOL_s00076g273 (76g273) and AOL_s00076g274 (76g274) and the transcription of 76g274, but not 76g273, was enhanced in X5. 76g274p had two conserved domains and was predicted as a nucleoprotein, which we confirmed by its nuclear localization in Saccharomyces cerevisiae using the green fluorescent protein-fused 76g274p. The transcription of 76g274 was stimulated or inhibited by several environmental factors. The sporulation yields of 76g274-deficient mutants were decreased by 70%, and transcription of several sporulation-related genes was severely diminished compared to the WT during the conidiation. In summary, a method for screening mutants was established in A. oligospora and using the method, we identified a novel C2H2-type transcription factor that positively regulates the conidiation of A. oligospora.

Brucella melitensis M5-90Δbp26 as a potential live vaccine that allows for the distinction between natural infection and immunization.

Brucella is Gram-negative intracellular bacterial pathogen that infects humans and animals and contributes to great economic losses in developing countries. Presently, live attenuated Brucella vaccines (Brucella melitensis M5-90) are the most effective means of brucellosis control and prevention in animals. However, these vaccines have several drawbacks, such as an inability to distinguish between a natural infection and immunization and an association with abortions in pregnant animals. Therefore, this study constructed a Brucella M5-90Δbp26 mutant and evaluated its virulence. The survival of the M5-90Δbp26 mutant was attenuated in human placenta trophoblastic 8 cells (HPT-8 cells) and in BALB/c mice, with a high immunoprotectivity noted in mice. Furthermore, safety tests showed that the M5-90Δbp26 mutant was less virulent than the M5-90 vaccine strain. Additionally, an indirect enzyme-linked immunosorbent assay (ELISA) screening was shown to detect the presence of Brucella protein 26 (BP26) with high sensitivity, with M5-90Δbp26 inoculation accompanied with a lack of BP26 expression, and was further confirmed by western blotting. Together, the M5-90Δbp26 mutant and the indirect ELISA can be employed to distinguish vaccinated livestock from infected animals.