Brain and blood transcriptome profiles delineate common genetic pathways across suicidal ideation and suicide.

Human genetic studies indicate that suicidal ideation and behavior are both heritable. Most studies have examined associations between aberrant gene expression and suicide behavior, but behavior risk is linked to the severity of suicidal ideation. Through a gene network approach, this study investigates how gene co-expression patterns are associated with suicidal ideation and severity using RNA-seq data in peripheral blood from 46 live participants with elevated suicidal ideation and 46 with no ideation. Associations with the presence of suicidal ideation were found within 18 co-expressed modules (p < 0.05), as well as in 3 co-expressed modules associated with suicidal ideation severity (p < 0.05, not explained by severity of depression). Suicidal ideation presence and severity-related gene modules with enrichment of genes involved in defense against microbial infection, inflammation, and adaptive immune response were identified and investigated using RNA-seq data from postmortem brain that revealed gene expression differences with moderate effect sizes in suicide decedents vs. non-suicides in white matter, but not gray matter. Findings support a role of brain and peripheral blood inflammation in suicide risk, showing that suicidal ideation presence and severity are associated with an inflammatory signature detectable in blood and brain, indicating a biological continuity between ideation and suicidal behavior that may underlie a common heritability.

Association of Blast Exposure in Military Breaching with Intestinal Permeability Blood Biomarkers Associated with Leaky Gut.

Injuries and subclinical effects from exposure to blasts are of significant concern in military operational settings, including tactical training, and are associated with self-reported concussion-like symptomology and physiological changes such as increased intestinal permeability (IP), which was investigated in this study. Time-series gene expression and IP biomarker data were generated from "breachers" exposed to controlled, low-level explosive blast during training. Samples from 30 male participants at pre-, post-, and follow-up blast exposure the next day were assayed via RNA-seq and ELISA. A battery of symptom data was also collected at each of these time points that acutely showed elevated symptom reporting related to headache, concentration, dizziness, and taking longer to think, dissipating ~16 h following blast exposure. Evidence for bacterial translocation into circulation following blast exposure was detected by significant stepwise increase in microbial diversity (measured via alpha-diversity p = 0.049). Alterations in levels of IP protein biomarkers (i.e., Zonulin, LBP, Claudin-3, I-FABP) assessed in a subset of these participants (n = 23) further evidenced blast exposure associates with IP. The observed symptom profile was consistent with mild traumatic brain injury and was further associated with changes in bacterial translocation and intestinal permeability, suggesting that IP may be linked to a decrease in cognitive functioning. These preliminary findings show for the first time within real-world military operational settings that exposures to blast can contribute to IP.

A genome-wide association study of suicidal behavior.

Genome wide array studies have reported limited success in identifying genetic markers conferring risk for suicidal behavior (SB). This may be attributable to study designs with primary outcome other than SB. We performed a GWAS on suicides and cases with a history of nonfatal suicide attempts compared with psychiatric controls and healthy volunteers. A consortium of USA, Canadian and German teams assembled two groups of cases (suicide attempters and suicides, N = 577) and non-attempter psychiatric and healthy controls (N = 1,233). Logistic regression was used to test genotype-suicidal behavior association. The test was repeated separating suicide attempt and completed suicide as outcomes. No SNP reached genome-wide significance, but several SNPs within STK3, ADAMTS14, PSME2, and TBX20 genes reached P < 1 × 10(-5) . The top SNPs for the suicide attempt analysis included two from DPP10, one from CTNNA3 and one from STK32B. In the suicide analysis we found seven SNPs from the TBX20 gene in the top hits. Pathway analysis identified the following pathways: "Cellular Assembly and Organization," "Nervous System Development and Function," "Cell Death and Survival," "Immunological Disease," "Infectious Disease," and "Inflammatory Response." The top genes in the SB analysis did not overlap with those in the ideation analysis. No genome wide significant results suggest that susceptibility to SB has genetic risk factors with smaller effect sizes. The strongest candidate genes, ADAMTS14, and PSME2 (both linked to inflammatory response), STK3 (neuronal cell death), and TBX20 (brainstem motor neuron development), have not been previously reported in association with suicide and warrant further study.

MethylomeDB: a database of DNA methylation profiles of the brain.

MethylomeDB (http://epigenomics.columbia.edu/methylomedb/index.html) is a new database containing genome-wide brain DNA methylation profiles. DNA methylation is an important epigenetic mark in the mammalian brain. In human studies, aberrant DNA methylation alterations have been associated with various neurodevelopmental and neuropsychiatric disorders such as schizophrenia, and depression. In this database, we present methylation profiles of carefully selected non-psychiatric control, schizophrenia, and depression samples. We also include data on one mouse forebrain sample specimen to allow for cross-species comparisons. In addition to our DNA methylation data generated in-house, we have and will continue to include published DNA methylation data from other research groups with the focus on brain development and function. Users can view the methylation data at single-CpG resolution with the option of wiggle and microarray formats. They can also download methylation data for individual samples. MethylomeDB offers an important resource for research into brain function and behavior. It provides the first source of comprehensive brain methylome data, encompassing whole-genome DNA methylation profiles of human and mouse brain specimens that facilitate cross-species comparative epigenomic investigations, as well as investigations of schizophrenia and depression methylomes.

Self-supervised learning for medical image analysis: Discriminative, restorative, or adversarial?

Discriminative, restorative, and adversarial learning have proven beneficial for self-supervised learning schemes in computer vision and medical imaging. Existing efforts, however, fail to capitalize on the potentially synergistic effects these methods may offer in a ternary setup, which, we envision can significantly benefit deep semantic representation learning. Towards this end, we developed DiRA, the first framework that unites discriminative, restorative, and adversarial learning in a unified manner to collaboratively glean complementary visual information from unlabeled medical images for fine-grained semantic representation learning. Our extensive experiments demonstrate that DiRA: (1) encourages collaborative learning among three learning ingredients, resulting in more generalizable representation across organs, diseases, and modalities; (2) outperforms fully supervised ImageNet models and increases robustness in small data regimes, reducing annotation cost across multiple medical imaging applications; (3) learns fine-grained semantic representation, facilitating accurate lesion localization with only image-level annotation; (4) improves reusability of low/mid-level features; and (5) enhances restorative self-supervised approaches, revealing that DiRA is a general framework for united representation learning. Code and pretrained models are available at https://github.com/JLiangLab/DiRA.

Impact of an internet-based insomnia intervention on suicidal ideation and associated correlates in veterans at elevated suicide risk.

Improving public health approaches to suicide prevention requires scalable evidence-based interventions that can be easily disseminated. Given empirical data supporting the association between insomnia and suicide risk, internet-delivered insomnia interventions are promising candidates to meet this need. The purpose of this study was to examine whether an unguided internet-delivered cognitive-behavioral therapy for insomnia (iCBT-I) improved insomnia severity, suicidal ideation (SI), and suicide risk correlates (depression, post-traumatic stress disorder, anxiety, hostility, belongingness, hopelessness, agitation, irritability, concentration) in a sample of veterans. Secondary data analysis of Operation Enduring Freedom, Operation Iraqi Freedom, and Operation New Dawn veterans (n = 50) with clinically significant insomnia and elevated SI drawn from a larger randomized controlled trial (RCT) of an iCBT-I, Sleep Healthy Using the Internet (SHUTi). Two-sample t-tests or Wilcoxon rank sum tests were used to evaluate between-group differences (SHUTi vs. Insomnia Education Website control) in symptom improvement from baseline to post-intervention. SHUTi participants experienced a significant improvement in insomnia severity (P < .001; d = -1.08) and a non-significant with small (subthreshold medium) effect size reduction of SI (P = .17, d = 0.40), compared to control participants. Significant improvement in hopelessness was observed (medium effect size), with non-significant small to medium effect size reductions in most remaining suicide risk correlates. Self-administered iCBT-I was associated with improvements in insomnia severity in veterans at elevated risk for suicide. These preliminary findings suggest that SI and suicide risk correlates may improve following an iCBT-I intervention, demonstrating the need for future well-powered iCBT-I RCTs targeted for populations at elevated suicide risk.

In this secondary data analysis, we examined improvements in insomnia severity, suicidal ideation (SI), and suicide risk correlates in veterans with clinically significant insomnia and elevated SI drawn from a larger randomized controlled trial (RCT) examining an unguided internet-delivered cognitive-behavioral therapy for insomnia (iCBT-I). Veterans in the iCBT-I group experienced greater improvements in insomnia severity and hopelessness than veterans in the Insomnia Education Website control. Although between-group differences in SI and other suicide risk correlates were not statistically significant, effect sizes suggest that SI and symptoms of depression, irritability, concentration, post-traumatic stress disorder, and hostility may improve following iCBT-I intervention. These results suggest that digital and iCBT-I interventions may be especially powerful tools for use in suicide prevention among veterans but highlight the critical need for additional large-scale studies to examine suicide-specific mechanisms and outcomes to guide implementation efforts.

Chromatin and sequence features that define the fine and gross structure of genomic methylation patterns.

Abnormalities of genomic methylation patterns are lethal or cause disease, but the cues that normally designate CpG dinucleotides for methylation are poorly understood. We have developed a new method of methylation profiling that has single-CpG resolution and can address the methylation status of repeated sequences. We have used this method to determine the methylation status of >275 million CpG sites in human and mouse DNA from breast and brain tissues. Methylation density at most sequences was found to increase linearly with CpG density and to fall sharply at very high CpG densities, but transposons remained densely methylated even at higher CpG densities. The presence of histone H2A.Z and histone H3 di- or trimethylated at lysine 4 correlated strongly with unmethylated DNA and occurred primarily at promoter regions. We conclude that methylation is the default state of most CpG dinucleotides in the mammalian genome and that a combination of local dinucleotide frequencies, the interaction of repeated sequences, and the presence or absence of histone variants or modifications shields a population of CpG sites (most of which are in and around promoters) from DNA methyltransferases that lack intrinsic sequence specificity.

Correlation between religious coping and depression in cancer patients.

BACKGROUND: Cancer often progresses very rapidly and either leads to various complications or patients eventually die of the disease. One of important consequences of cancer is depression which can increase the morbidity and mortality in non-treated cases. Religious coping is the use of religious beliefs or practices to reduce distress and deal with problems in life. This study aimed to determine the relationship between religious coping and depression in cancer patients. SUBJECTS AND METHODS: A descriptive-correlational study was conducted on 150 consequent cancer patients in three centers: Imam-Reza Hospital in Birjand, Qaem and Omid hospitals in Mashhad. Two questionnaires including Pargament's questionnaire for evaluation of religious coping and the Beck depression inventory were used. Data analysis was performed using multiple regression and correlation. RESULTS: There was no significant difference between men and women in the mean score of avoidant relationship with God and alternate fearfulness and hopefulness (ambivalence coping style). But the mean score of relationship with God in women was higher than men. The rate of depression was higher among patients who had an avoidant strategy. The religious coping method of relationship with God was effective in reducing depression. The rate of depression was lower among patients whose families had a better attitude to religion. CONCLUSIONS: Psychotherapy, individual/familial counseling, and especially increasing of religious beliefs such as praying and trust in God, as well as increasing the knowledge of patient and his/her family cause better acceptance of the disease and better confrontation of psychological problems.

Assessing sleep health dimensions in frontline registered nurses during the COVID-19 pandemic: implications for psychological health and wellbeing.

The COVID-19 pandemic altered work environments of nurses, yielding high rates of stress and burnout. Potential protective factors, including effective sleep, may influence psychological health and wellbeing. Evidence about sleep in nurses may help develop interventions that mitigate burnout and poor psychological outcomes. A cross sectional survey was distributed across three hospitals to nurses in New York City (NYC). During the first wave of the pandemic (March-April 2020), NYC had the highest incidence of laboratory-confirmed COVID-19 cases (915/100 000) and half of all COVID-related deaths nationwide. Multivariable logistic regression was used to determine associations between Pittsburgh Sleep Quality Index (PSQI) global sleep score, PSQI sleep dimensions, and psychological health (burnout, depression, anxiety, and compassion fatigue), unadjusted and then controlling for individual and professional characteristics. More than half of the participants reported burnout (64%), depression, (67%), and anxiety (77%). Eighty percent of participants had PSQI global scores >5 (poor sleep) (mean 9.27, SD 4.14). Respondents reporting good sleep (PSQI ≤ 5) had over five times the odds of no burnout (OR: 5.65, 95% CI: 2.60, 12.27); increased odds of screening negative for depression (OR: 6.91, 95% CI: 3.24, 14.72), anxiety (OR: 10.75, 95% CI: 4.22, 27.42), and compassion fatigue (OR: 7.88, 95% CI: 1.97, 31.51). Poor subjective sleep quality PSQI subcomponent was associated with burnout (OR: 2.21, 95% CI: 1.41, 3.48) but sleep duration subcomponent was not (OR: 0.84, 95% CI: 0.59, 1.19). Daytime dysfunction was significantly associated with all psychological outcomes. Sleep disturbances and medications yielded higher anxiety odds. Overall, sleep quality appears more strongly related to burnout than sleep duration in nurses working during the COVID-19 pandemic. Sleep interventions should target individual sleep dimensions in nurses.

Delivery of Telehealth Complementary and Integrated Health Interventions Improves Mental Health and Overall Wellness to Broaden Veterans' Access to Care.

Background: Complementary and integrative health (CIH) interventions show promise in improving overall wellness and engaging Veterans at risk of suicide. Methods: An intensive 4-week telehealth CIH intervention programming was delivered motivated by the COVID-19 pandemic, and outcomes were measured pre-post program completion. Results: With 93% program completion (121 Veterans), significant reduction in depression and post-traumatic stress disorder symptoms were observed pre-post telehealth CIH programing, but not in sleep quality. Improvements in pain symptoms, and stress management skills were observed in Veterans at risk of suicide. Discussion: Telehealth CIH interventions show promise in improving mental health symptoms among at-risk Veterans, with great potential to broaden access to care toward suicide prevention.

Functional Architecture of Brain and Blood Transcriptome Delineate Biological Continuity Between Suicidal Ideation and Suicide.

Human genetic studies indicate that suicidal ideation and behavior are both heritable. Most studies have examined associations between aberrant gene expression and suicide behavior, but behavior risk is linked to severity of suicidal ideation. Through a gene network approach, this study investigates how gene co-expression patterns are associated with suicidal ideation and severity using RNA-seq data in peripheral blood from 46 live participants with elevated suicidal ideation and 46 with no ideation. Associations with presence and severity of suicidal ideation were found within 18 and 3 co-expressed modules respectively (p < 0.05), not explained by severity of depression. Suicidal ideation presence and severity-related gene modules with enrichment of genes involved in defense against microbial infection, inflammation, and adaptive immune response were identified, and tested using RNA-seq data from postmortem brain that revealed gene expression differences in suicide decedents vs. non-suicides in white matter, but not gray matter. Findings support a role of brain and peripheral blood inflammation in suicide risk, showing that suicidal ideation presence and severity is associated with an inflammatory signature detectable in blood and brain, indicating a biological continuity between ideation and suicidal behavior that may underlie a common heritability.

The relationship of myocardial and liver T2* values with cardiac function and laboratory findings in transfusion-dependent thalassemia major patients: A retrospective cardiac MRI study.

Introduction: Cardiac complications are the leading cause of death in thalassemia patients. It is assumed that progressive iron accumulation results in myocyte damage. Myocardial T2* measurement by cardiac MRI quantifies iron overload. We aimed to study the association between left and right ventricular (LV and RV) function and iron deposition estimation by cardiac MRI T2* in a sample of Iranian patients. Methods: Cardiac MRI exams of 118 transfusion-dependent thalassemia major patients were evaluated retrospectively. Biventricular function and volume and myocardial and liver T2* values were measured. The demographic and lab data were registered. Poisson and chi-square regression analyses investigated the correlation between the T2* value and ventricular dysfunction. Results: The study participants' mean (SD) age was 32.7y (9.02), and 47.46% were female. Forty-nine cases (41.52%) revealed at least uni-ventricular dysfunction. LV dysfunction was noted in 20 cases, whereas 47 patients revealed RV dysfunction. The risk of LV dysfunction was 5.3-fold higher in patients with cardiac T2* value less than 10msec (RR=5.3, 95% CI=1.6, 17.1, P<0.05). No association was found between age, liver T2* value, serum ferritin level, and chelation therapy with the risk of LV and RV dysfunction. Conclusion: Cardiac MRI T2* measure is a good indicator of LV dysfunction. Moreover, MRI parameters, especially RV functional measures, may have a substantial role in patient management. Therefore, cardiac MRI should be included in beta-thalassemia patients' management strategies.

A cross-sectional study of correlations among blood-based biomarkers for intestinal permeability: A pilot study of United States veterans with posttraumatic stress disorder symptoms.

While many studies of intestinal permeability (IP) are focused on those with gastrointestinal (GI) disorders, there is a rising trend to analyze IP among individuals with mental health conditions including posttraumatic stress disorder (PTSD) with and without diagnosed GI conditions. This interest stems from the association between gut dysbiosis and chronic inflammation, which are mechanisms linked to stress-related somatic and mental health conditions. Efforts to date have resulted in the exploration of non-invasive and feasible measures to identify an IP biomarker that could also serve as a treatment target. Additionally, those conducting studies regarding IP often recruit individuals without health problems and compare levels of biomarkers of IP to those obtained from participants with conditions of interest. This study aimed to assess correlations between blood-based biomarkers of IP, as well as examine the association between blood-based biomarkers of IP and PTSD symptoms. Blood was sampled from seventeen United States military Veterans with variable severity of PTSD symptoms per the posttraumatic checklist for DSM-5 (PCL-5) (n = 6 with scores over 31 indicating clinically meaningful symptoms of PTSD; overall range 0-49, mean 20.8, standard deviation 15.7) and analyzed blood biomarkers of IP including citrulline, diamine oxidase, glucagon-like peptide-2, intestinal fatty-acid binding protein, lipopolysaccharide binding protein, lipopolysaccharide, and zonulin. Correlations between the IP blood-based biomarkers ranged from ρ of -0.31 to 0.35. None of the measured biomarkers were significantly correlated to PTSD symptom severity scores (ρ of -0.34 to 0.05). Although based on limited sample size, our results call into question the specificity of blood-based biomarkers of IP when: (1) studying persons with and without PTSD symptoms in whom clinical GI disorders are not necessarily the focus of the study; and (2) comparing IP results among individuals with well-defined disease states to those without the disease (e.g., controls). Further studies are needed to explore the role of external factors (e.g., nutrition, obesity, alcohol use) on IP and to determine if the biomarkers studied are appropriate for measuring IP in people with a range of symptoms related to PTSD.

Association of interleukin-6 with suicidal ideation in veterans: a longitudinal perspective.

Introduction: Studies showing associations between inflammation in suicide are typically cross-sectional. Present study investigated how cytokine levels track with suicidal ideation and severity longitudinally. Methods: Veterans with a diagnosis of major depressive disorder (MDD) with or without suicide attempt history (MDD/SA n = 38, MDD/NS n = 41) and non-psychiatric non-attempter controls (HC n = 33) were recruited, MDD/SA and HC groups were followed longitudinally at 3 months and 6 months. Blood plasma was collected and processed using Luminex Immunology Multiplex technology. Results: Significant differences in depression severity (BDI) and suicidal ideation severity (SSI) were observed across all groups at study entry, wherein MDD/SA group had the highest scores followed by MDD/NS and HC, respectively. Cytokines IL-1ß, IL-4, TNF-α, IFN-γ, and IL-6 were examined at study entry and longitudinally, with IL6 levels differing significantly across the groups (p = 0.0123) at study entry. Significant differences in changes in cytokine levels between depressed attempters and the control group were detected for IL-6 (interaction F1,91.77 = 5.58, p = 0.0203) and TNF-α (F1,101.73 = 4.69, p = 0.0327). However, only depressed attempters showed a significant change, in IL-6 and TNF-α levels, decreasing over time [IL-6: b = -0.04, 95% CI = (-0.08, -0.01), p = 0.0245 and TNF-α: b = -0.02, 95% CI = (-0.04, -0.01), p = 0.0196]. Although IL-6 levels were not predictive of suicidal ideation presence [OR = 1.34, 95% CI = (0.77, 2.33), p = 0.3067], IL-6 levels were significantly associated with suicidal ideation severity (b = 0.19, p = 0.0422). Discussion: IL-6 was not associated with presence of suicidal ideation. IL-6 however, was associated with severity of ideation, suggesting that IL-6 may be useful in clinical practice, as an objective marker of heightened suicide risk.

Smaller rostral cingulate volume and psychosocial correlates in veterans at risk for suicide.

Suicide research/clinical work remain in dire need of effective tools that can better predict suicidal behavior. A growing body of literature has started to focus on the role that neuroimaging may play in helping explain the path towards suicide. Specifically, structural alterations of rostral anterior cingulate cortex (rost-ACC) may represent a biological marker and/or indicator of suicide risk in Major Depressive Disorder (MDD). Furthermore, the construct of "grit," defined as perseverance for goal-attainment and shown to be associated with suicidality, is modulated by rost-ACC. The aim was to examine relationships among rost-ACC gray matter volume, grit, and suicidality in U.S. Military Veterans. Participants were age-and-sex-matched Veterans with MDD: with suicide attempt (MDD+SA:n = 23) and without (MDD-SA:n = 37). Groups did not differ in depression symptomatology. Participants underwent diagnostic interview, clinical symptom assessment, and 3T-MRI-scan. A Group (SA-vs.-No-SA) x Cingulate-region (rostral-caudal-posterior) x Hemisphere (left-right) mixed-model-multivariate-ANOVA was conducted. Left-rost-ACC was significantly smaller in MDD+SA, Group x Cingulate-region x Hemisphere-interaction. Lower grit and less left-rost-ACC gray matter each predicted suicide attempt history, but grit level was a more robust predictor of SA. Both structural alterations of rost-ACC and grit level represent potentially valuable tools for suicide risk assessment.

Methyl-Analyzer--whole genome DNA methylation profiling.

SUMMARY: Methyl-Analyzer is a python package that analyzes genome-wide DNA methylation data produced by the Methyl-MAPS (methylation mapping analysis by paired-end sequencing) method. Methyl-MAPS is an enzymatic-based method that uses both methylation-sensitive and -dependent enzymes covering >80% of CpG dinucleotides within mammalian genomes. It combines enzymatic-based approaches with high-throughput next-generation sequencing technology to provide whole genome DNA methylation profiles. Methyl-Analyzer processes and integrates sequencing reads from methylated and unmethylated compartments and estimates CpG methylation probabilities at single base resolution. AVAILABILITY AND IMPLEMENTATION: Methyl-Analyzer is available at http://github.com/epigenomics/methylmaps. Sample dataset is available for download at http://epigenomicspub.columbia.edu/methylanalyzer_data.html. CONTACT: fgh3@columbia.edu SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Correction to "Does the repeat dose of gonadotropin-releasing hormone agonist trigger in polycystic ovarian syndrome improve in vitro fertilization cycles outcome? A clinical trial study" [Int J Reprod BioMed 2020; 18: 485-490].

[This corrects the article on p. 485 in vol. 18 PMCPMC7385917.].

Diagnostic Value of Immunoglobulin G Anti-Deamidated Gliadin Peptide Antibody for Diagnosis of Pediatric Celiac Disease: A Study from Shiraz, Iran.

Purpose: Screening serologic tests are important tools for the diagnosis of celiac disease (CD). Immunoglobulin (Ig)G anti-deamidated gliadin peptide (anti-DGP) is a relatively new autoantibody thought to have good diagnostic accuracy, comparable to that of anti-tissue transglutaminase (anti-tTG) antibody. Methods: Pediatric patients (n=86) with a clinical suspicion of CD were included. Duodenal biopsy, anti-tTG, and IgG anti-DGP antibody tests were performed. The patients were divided into CD and control groups based on the pathological evaluation of duodenal biopsies. The diagnostic accuracy of serological tests was determined. Results: IgA anti-tTG and IgG anti-DGP antibodies were positive in 86.3% and 95.4% of patients, respectively. The sensitivity, specificity, and diagnostic accuracy of the IgA anti-tTG test were 86.3%, 50.0%, and 68.6%, respectively, and those of the IgG anti-DGP test were 95.4%, 85.7%, and 90.7%, respectively. The area under the receiver operating characteristic (ROC) curve was 0.84 (95% confidence interval [CI], 0.74-0.91) for IgA anti-tTG test and 0.93 (95% CI, 0.86-0.97) for IgG anti-DGP test. The comparison of IgA anti-tTG and IgG anti-DGP ROC curves showed a higher sensitivity and specificity of the IgG anti-DGP test. Conclusion: IgG anti-DGP is a reliable serological test for CD diagnosis in children. High tTG and DGP titers in the serum are suggestive of severe duodenal atrophy. The combined use of IgA anti-tTG and IgG anti-DGP tests for the initial screening of CD can improve diagnostic sensitivity.

CAiD: Context-Aware Instance Discrimination for Self-supervised Learning in Medical Imaging.

Recently, self-supervised instance discrimination methods have achieved significant success in learning visual representations from unlabeled photographic images. However, given the marked differences between photographic and medical images, the efficacy of instance-based objectives, focusing on learning the most discriminative global features in the image (i.e., wheels in bicycle), remains unknown in medical imaging. Our preliminary analysis showed that high global similarity of medical images in terms of anatomy hampers instance discrimination methods for capturing a set of distinct features, negatively impacting their performance on medical downstream tasks. To alleviate this limitation, we have developed a simple yet effective self-supervised framework, called Context-Aware instance Discrimination (CAiD). CAiD aims to improve instance discrimination learning by providing finer and more discriminative information encoded from a diverse local context of unlabeled medical images. We conduct a systematic analysis to investigate the utility of the learned features from a three-pronged perspective: (i) generalizability and transferability, (ii) separability in the embedding space, and (iii) reusability. Our extensive experiments demonstrate that CAiD (1) enriches representations learned from existing instance discrimination methods; (2) delivers more discriminative features by adequately capturing finer contextual information from individual medial images; and (3) improves reusability of low/mid-level features compared to standard instance discriminative methods. As open science, all codes and pre-trained models are available on our GitHub page: https://github.com/JLiangLab/CAiD.

DiRA: Discriminative, Restorative, and Adversarial Learning for Self-supervised Medical Image Analysis.

Discriminative learning, restorative learning, and adversarial learning have proven beneficial for self-supervised learning schemes in computer vision and medical imaging. Existing efforts, however, omit their synergistic effects on each other in a ternary setup, which, we envision, can significantly benefit deep semantic representation learning. To realize this vision, we have developed DiRA, the first framework that unites discriminative, restorative, and adversarial learning in a unified manner to collaboratively glean complementary visual information from unlabeled medical images for fine-grained semantic representation learning. Our extensive experiments demonstrate that DiRA (1) encourages collaborative learning among three learning ingredients, resulting in more generalizable representation across organs, diseases, and modalities; (2) outperforms fully supervised ImageNet models and increases robustness in small data regimes, reducing annotation cost across multiple medical imaging applications; (3) learns fine-grained semantic representation, facilitating accurate lesion localization with only image-level annotation; and (4) enhances state-of-the-art restorative approaches, revealing that DiRA is a general mechanism for united representation learning. All code and pretrained models are available at https://github.com/JLiangLab/DiRA.