Clinical Datasets Research Articles

Abstract PO-031: Systematic discovery of chemotherapy response biomarkers in Peripheral T Cell Lymphomas

Abstract Background. Identifying which patients will benefit from individual therapies remains a general challenge in oncology. Predictive biomarkers of drug sensitivity, when known, can increase the rate of success of trials, and allow stratification of patients for targeted combination therapies, as recent shown by the GUIDANCE-01 trial in Large B-Cell Lymphoma. Peripheral T-Cell Lymphomas (PTCL) could particularly benefit from this approach, being heterogeneous diseases against which many therapies are active in unpredictable subsets of patients. A central challenge of biomarker discovery is finding true determinants of drug response, which may be multiple genes, among thousands of molecular features that may accidentally correlate. Here, we aim to mitigate this challenge by combining functional genetic screens and gene expression profiling in diverse PTCL cultures, to enable the systematic discovery of drug sensitivity biomarkers. Methods. To discover genes whose expression directly modifies drug sensitivity, we conducted genome-wide CRISPR screens with PTCL cells under treatment with five therapies with clinical activity in PTCL: Romidepsin, Pralatrexate, Gemcitabine, Oxaliplatin, and Bortezomib. To discover genes whose basal expression significantly correlates with specific drug sensitivities, we measured RNAseq and drug response functions for a diverse panel of 27 PTCL cultures. Genes in the intersection of both experiments are causally related to and correlate with drug responses in PTCL, and have the statistical benefit of detection by independent methods. Results. Functional genetic screens on therapy provided mechanistic insights for each drug of interest, and identified genes whose perturbation affects drug sensitivity. Whereas hundreds of genes nominally correlate with drug sensitivity in cell line panels, for each drug a shortlist of only 5 to 20 genes are both CRISPR hits and correlate with response across PTCL cultures. For each drug tested, multi-gene sensitivity scores constructed from hits stratify PTCL cultures into statistically distinct sensitive and non-sensitive subgroups (Mann Whitney P-Values <0.01). These genes’ relationships to drug mechanisms were often supported by prior studies, including functions in DNA repair, cell cycle regulation, and solute carriers required for cellular entry of specific therapies, such as SLC19A1 for Pralatrexate and SLC43A2 for Oxaliplatin. Conclusions. Combining functional genetic screens and correlative drug sensitivity measurements narrowed the search for predictive biomarkers of drug sensitivity, and may be a general means of refining biomarker hypotheses for validation in clinical datasets. Although chemotherapy sensitivity is commonly polygenic, multi-gene scores based on few mechanistically-related genes were able to significantly distinguish drug sensitive from resistant PTCLs. These approaches provide novel hypotheses for drug-gene relationships that may serve as the basis for molecular subtyping of individual lymphomas for precision regimens. Citation Format: Adam C Palmer, Jacob C Pantazis. Systematic discovery of chemotherapy response biomarkers in Peripheral T Cell Lymphomas [abstract]. In: Proceedings of the Fourth AACR International Meeting on Advances in Malignant Lymphoma: Maximizing the Basic-Translational Interface for Clinical Application; 2024 Jun 19-22; Philadelphia, PA. Philadelphia (PA): AACR; Blood Cancer Discov 2024;5(3_Suppl):Abstract nr PO-031.

Epidemiological trends and factors associated with survival in patients with medulloblastoma: A 45-year population-based retrospective study

Medulloblastoma (MB) is a primary brain malignancy. However, updated epidemiological data and long-term outcomes are lacking.The clinical and epidemiological datasets of patients with MB in the current study were obtained from the Surveillance, Epidemiology, and End Results (SEER) databases. Joinpoint regression models were used to assess the rate of changes in the incidence, prevalence, and treatment trends in patients with MB. Cox hazard and competition risk model analyses were used to assess overall survival (OS) and cancer-specific survival (CSS).The age-adjusted incidence of MB remained relatively stable at 0.15 per 100,000 individuals in 2019. The annual percentage change (APC) of females remained stable, whereas that of males increased over time. The 20-year limited-duration prevalence of patients with MB increased significantly from 0.00016 % in 1999 to 0.00203 % in 2018. Patients aged 5–19 years accounted for 46.7 % of all age groups, and the trend for the three treatments was increased. Average annual percentage change (AAPC) for the chemotherapy group was increased in patients aged 20 + years MB [AAPC = 2.66 (95 % CI 0.93–6.31)]. Multivariate analysis revealed that OS and CSS varied significantly according to age, year of diagnosis, histology, stage, surgery, and radiotherapy. Subgroup analysis showed that chemotherapy was associated with a favorable prognosis in high-risk groups.The incidence of MB remained relatively stable, and its prevalence increased significantly. This current population-based study further identified the prognostic factors in patients with MB. Moreover, the use of chemotherapy was associated with better survival in high-risk groups.

An integrated single-cell RNA-seq map of human neuroblastoma tumors and preclinical models uncovers divergent mesenchymal-like gene expression programs

BackgroundNeuroblastoma is a common pediatric cancer, where preclinical studies suggest that a mesenchymal-like gene expression program contributes to chemotherapy resistance. However, clinical outcomes remain poor, implying we need a better understanding of the relationship between patient tumor heterogeneity and preclinical models.ResultsHere, we generate single-cell RNA-seq maps of neuroblastoma cell lines, patient-derived xenograft models (PDX), and a genetically engineered mouse model (GEMM). We develop an unsupervised machine learning approach (“automatic consensus nonnegative matrix factorization” (acNMF)) to compare the gene expression programs found in preclinical models to a large cohort of patient tumors. We confirm a weakly expressed, mesenchymal-like program in otherwise adrenergic cancer cells in some pre-treated high-risk patient tumors, but this appears distinct from the presumptive drug-resistance mesenchymal programs evident in cell lines. Surprisingly, however, this weak-mesenchymal-like program is maintained in PDX and could be chemotherapy-induced in our GEMM after only 24 h, suggesting an uncharacterized therapy-escape mechanism.ConclusionsCollectively, our findings improve the understanding of how neuroblastoma patient tumor heterogeneity is reflected in preclinical models, provides a comprehensive integrated resource, and a generalizable set of computational methodologies for the joint analysis of clinical and pre-clinical single-cell RNA-seq datasets.

Comparison of Synthetic Data Generation Techniques for Control Group Survival Data in Oncology Clinical Trials: Simulation Study.

Synthetic patient data (SPD) generation for survival analysis in oncology trials holds significant potential for accelerating clinical development. Various machine learning methods, including classification and regression trees (CART), random forest (RF), Bayesian network (BN), and conditional tabular generative adversarial network (CTGAN), have been used for this purpose, but their performance in reflecting actual patient survival data remains under investigation. The aim of this study was to determine the most suitable SPD generation method for oncology trials, specifically focusing on both progression-free survival (PFS) and overall survival (OS), which are the primary evaluation end points in oncology trials. To achieve this goal, we conducted a comparative simulation of 4 generation methods, including CART, RF, BN, and the CTGAN, and the performance of each method was evaluated. Using multiple clinical trial data sets, 1000 data sets were generated by using each method for each clinical trial data set and evaluated as follows: (1) median survival time (MST) of PFS and OS; (2) hazard ratio distance (HRD), which indicates the similarity between the actual survival function and a synthetic survival function; and (3) visual analysis of Kaplan-Meier (KM) plots. Each method's ability to mimic the statistical properties of real patient data was evaluated from these multiple angles. In most simulation cases, CART demonstrated the high percentages of MSTs for synthetic data falling within the 95% CI range of the MST of the actual data. These percentages ranged from 88.8% to 98.0% for PFS and from 60.8% to 96.1% for OS. In the evaluation of HRD, CART revealed that HRD values were concentrated at approximately 0.9. Conversely, for the other methods, no consistent trend was observed for either PFS or OS. CART demonstrated better similarity than RF, in that CART caused overfitting and RF (a kind of ensemble learning approach) prevented it. In SPD generation, the statistical properties close to the actual data should be the focus, not a well-generalized prediction model. Both the BN and CTGAN methods cannot accurately reflect the statistical properties of the actual data because small data sets are not suitable. As a method for generating SPD for survival data from small data sets, such as clinical trial data, CART demonstrated to be the most effective method compared to RF, BN, and CTGAN. Additionally, it is possible to improve CART-based generation methods by incorporating feature engineering and other methods in future work.

Oral pimonidazole unveils clinicopathologic and epigenetic features of hypoxic tumour aggressiveness in localized prostate cancer

BackgroundTumor hypoxia is associated with prostate cancer (PCa) treatment resistance and poor prognosis. Pimonidazole (PIMO) is an investigational hypoxia probe used in clinical trials. A better understanding of the clinical significance and molecular alterations underpinning PIMO-labeled tumor hypoxia is needed for future clinical application. Here, we investigated the clinical significance and molecular alterations underpinning PIMO-labeled tumor hypoxia in patients with localized PCa, in order to apply PIMO as a prognostic tool and to identify potential biomarkers for future clinical translation.MethodsA total of 39 patients with localized PCa were recruited and administered oral PIMO before undergoing radical prostatectomy (RadP). Immunohistochemical staining for PIMO was performed on 37 prostatectomy specimens with staining patterns evaluated and clinical association analyzed. Whole genome bisulfite sequencing was performed using laser-capture of microdissected specimen sections comparing PIMO positive and negative tumor areas. A hypoxia related methylation molecular signature was generated by integrating the differentially methylated regions with previously established RNA-seq datasets.ResultsThree PIMO staining patterns were distinguished: diffuse, focal, and comedo-like. The comedo-like staining pattern was more commonly associated with adverse pathology. PIMO-defined hypoxia intensity was positively correlated with advanced pathologic stage, tumor invasion, and cribriform and intraductal carcinoma morphology. The generated DNA methylation signature was found to be a robust hypoxia biomarker, which could risk-stratify PCa patients across multiple clinical datasets, as well as be applicable in other cancer types.ConclusionsOral PIMO unveiled clinicopathologic features of disease aggressiveness in localized PCa. The generated DNA methylation signature is a novel and robust hypoxia biomarker that has the potential for future clinical translation.

Decoding the Clavien-Dindo Classification: Artificial Intelligence (AI) as a Novel Tool to Grade Postoperative Complications.

To assess ChatGPT's capability of grading postoperative complications using the Clavien-Dindo classification (CDC) via Artificial Intelligence (AI) with Natural Language Processing (NLP). The CDC standardizes grading of postoperative complications. However, consistent, and precise application in dynamic clinical settings is challenging. AI offers a potential solution for efficient automated grading. ChatGPT's accuracy in defining the CDC, generating clinical examples, grading complications from existing scenarios, and interpreting complications from fictional clinical summaries, was tested. ChatGPT 4 precisely mirrored the CDC, outperforming version 3.5. In generating clinical examples, ChatGPT 4 showcased 99% agreement with minor errors in urinary catheterization. For single complications, it achieved 97% accuracy. ChatGPT was able to accurately extract, grade, and analyze complications from free text fictional discharge summaries. It demonstrated near perfect performance when confronted with real-world discharge summaries: comparison between the human and ChatGPT4 grading showed a κ value of 0.92 (95% CI 0.82-1) (P<0.001). ChatGPT 4 demonstrates promising proficiency and accuracy in applying the CDC. In the future, AI has the potential to become the mainstay tool to accurately capture, extract, and analyze CDC data from clinical datasets.

Significance of Multi-Cancer Genome Profiling Testing for Breast Cancer: A Retrospective Analysis of 3326 Cases from Japan's National Database.

Background: Breast cancer (BC) has the highest morbidity rate and the second-highest mortality rate of all cancers among women. Recently, multi-cancer genome profiling (multi-CGP) tests have become clinically available. In this study, we aimed to clarify the significance of multi-CGP testing of BC by using the large clinical dataset from The Center for Cancer Genomics and Advanced Therapeutics (C-CAT) profiling database in Japan. Materials and Methods: A total of 3744 BC cases were extracted from the C-CAT database, which enrolled 60,250 patients between June 2019 and October 2023. Of the 3744 BC cases, a total of 3326 cases for which the C-CAT included information on ER, PR, and HER2 status were classified into four subtypes, including TNBC, HR+/HER2-, HR+/HER2+, and HR-/HER2+. Comparisons between groups were performed by the χ2 test or Fisher's exact test using EZR. Kaplan-Meier curves were created using the log-rank test. Results: Of all 3326 cases analyzed, 1114 (33.5%) were TNBC cases, HR+/HER2- accounted for 1787 cases (53.7%), HR+/HER2+ for 260 cases (7.8%), and HR-/HER2+ for 165 cases (5.0%). Genetic abnormalities were most frequently detected in TP53 (58.0%), PIK3CA (35.5%), MYC (18.7%), FGF19 (15.5%), and GATA3 (15.1%) across all BCs. The rate of TMB-High was 12.3%, and the rate of MSI-High was 0.3%, in all BC cases. Therapeutic drugs were proposed for patients with mutations in six genes: PIK3CA, ERBB2, PTEN, FGFR1, ESR1, and AKT1. The prognoses of HR+/HER2- cases were significantly (p = 0.044) better in the treated group than in the untreated group. Conclusions: These findings suggest that cancer gene panel testing is useful for HR+/HER2- cases.

Leveraging a clinical emergency department dataset to estimate two-dose COVID-19 vaccine effectiveness and duration of protection in Canada

BackgroundDuring the COVID-19 pandemic, clinical care shifted toward virtual and Emergency Department care. We explored the feasibility of mRNA vaccine effectiveness (VE) estimation against SARS-CoV-2-related Emergency Department visits and hospitalizations using prospectively collected Emergency Department data. MethodsWe estimated two-dose VE using a test-negative design and data from 10 participating sites of the Canadian COVID-19 Emergency Department Rapid Response Network (CCEDRRN). We included Emergency Department patients presenting with COVID-19 symptoms and nucleic acid amplification testing for SARS-CoV-2 between July 19 and December 31, 2021. We excluded patients with unclear vaccination and one or more than 2 vaccine doses by their Emergency Department visit. ResultsAmong 3,405 eligible patients, adjusted two-dose mRNA VE against SARS-CoV-2-related Emergency Department visits was 93.3 % (95 % CI 87.9–96.3 %) between 7–55 days, sustained over 80 % through 139 days post-vaccination. In stratified analyses, VE was similar among patients with select immune-compromising conditions, chronic kidney disease, lung disease, unstable housing, and reported illicit substance use. ConclusionsTwo-dose mRNA VE against SARS-CoV-2-related Emergency Department visit was high and sustained, including among vulnerable subgroups. Compared to administrative datasets, active Emergency Department enrolment enables standardization for testing access and indication and supports separate VE assessment among special population subgroups. Compared to other active enrolment settings, Emergency Departments more consistently function during crises when alternate healthcare sectors become variably closed.Trial registration: Clinicaltrials.gov, NCT0470294.

Networking Aspects of the Electronic Health Records: Hypertext Transfer Protocol Version 2 (HTTP/2) vs HTTP/3.

The rapid development of the digital healthcare and the electronic health records (EHR) requires smooth networking infrastructure to access data using Hypertext Transfer Protocol (HTTP)-based applications. The new HTTP/3 standard should provide performance and security improvements over HTTP/2. The goal of our work was to test the performance of HTTP/2 and HTTP/3 in the context of the EHRs. We used 45,000 test FHIR Patient resources downloaded and uploaded using 20, 50, 100 and 200 resources per Bundle, which resulted in 2251, 901, 451 and 226 HTTP GET and POST requests respectively. The first test downloading 20 resources per Bundle showed that HTTP/3 outperformed HTTP/2 in the local (mean request time 16.57 ms ± 7.2 standard deviation [SD]) and in the remote network (71.45 ms ± 43.5 SD) which is almost 3 times faster. In the 50 and 100 resources per Bundle test the HTTP/3 protocol demonstrated again more than two times gain in downloading performance for remote requests with mean request time 91.13 ms ± 34.54 SD and 88.09 ms ± 21.66 SD respectively. Furthermore, HTTP/3 outperformed HTTP/2 in the constructed clinical dataset remote transfer. In the upload tests HTTP/3 showed only a slight gain in performance merely in the remote network. The HTTP/3 protocol is a relatively new development and a major improvement for the worldwide web. This new technology is still missing in the digital health and EHRs. Its use could offer a major performance gain in situations where data is gathered from multiple remote locations.

Gynecological Laparoscopic Surgeries under Spinal Anesthesia: Benefits and Challenges.

This prospective study investigated the feasibility of performing laparoscopic pelvic surgery under spinal anesthesia and analyzed the intraoperative side effects, like pain, nausea, and vomitus, of 915 patients. The implementation and performance of laparoscopic surgery under local anesthesia on 915 patients (out of a total of 3212 who underwent laparoscopic pelvic surgery under spinal anesthesia) were analyzed in relation to BMI (body mass index), obesity, pain during surgery, amount of intraperitoneal mmHg CO2 gas pressure, and surgical complications. BMI > 30, intra-abdominal adhesions, increased duration of the operation, bleeding, and increased intraperitoneal CO2 pressure were statistically significant as the main causes of pain during laparoscopic surgery under spinal anesthesia. Underweight patients, on the other hand, had less pain when intra-abdominal pressure increased compared to those of normal weight. The appearance of pain, nausea, and vomitus occurred in 10.3% of patients, and these events were easy to manage and treat. They did not affect the surgeon's work or the course of the operation. In light of these observations, we are proposing spinal anesthesia for laparoscopic surgery as the first choice in patients who have no contraindications. To the best of our knowledge, this clinical study constitutes the largest clinical observation and dataset concerning spinal anesthesia in laparoscopic pelvic surgery. ISRCTN38987, 10 December 2019.

DBGAN: Dual Branch Generative Adversarial Network for Multi-Modal MRI Translation

Existing magnetic resonance imaging translation models rely on generative adversarial networks, primarily employing simple convolutional neural networks. Unfortunately, these networks struggle to capture global representations and contextual relationships within magnetic resonance images. While the advent of Transformers enables capturing long-range feature dependencies, they often compromise the preservation of local feature details. To address these limitations and enhance both local and global representations, we introduce DBGAN , a novel dual-branch generative adversarial network. In this framework, the Transformer branch comprises sparse attention blocks and dense self-attention blocks, allowing for a wider receptive field while simultaneously capturing local and global information. The convolutional neural network branch, built with integrated residual convolutional layers, enhances local modeling capabilities. Additionally, we propose a fusion module that cleverly integrates features extracted from both branches. Extensive experimentation on two public datasets and one clinical dataset validates significant performance improvements with DBGAN. On Brats2018, it achieves a 10% improvement in MAE, 3.2% in PSNR, and 4.8% in SSIM for image generation tasks compared to RegGAN. Notably, the generated MRIs receive positive feedback from radiologists, underscoring the potential of our proposed method as a valuable tool in clinical settings.

A proteomic approach supports the clinical relevance of TAT-Cx43266-283 in glioblastoma

Glioblastoma (GBM) is the most frequent and aggressive primary brain cancer. The Src inhibitor, TAT-Cx43266-283, exerts antitumor effects in in vitro and in vivo models of GBM. Because addressing the mechanism of action is essential to translate these results to a clinical setting, in this study we carried out an unbiased proteomic approach. Data-independent acquisition mass spectrometry proteomics allowed the identification of 190 proteins whose abundance was modified by TAT-Cx43266-283. Our results were consistent with the inhibition of Src as the mechanism of action of TAT-Cx43266-283 and unveiled antitumor effectors, such as p120 catenin. Changes in the abundance of several proteins suggested that TAT-Cx43266-283 may also impact the brain microenvironment. Importantly, the proteins whose abundance was reduced by TAT-Cx43266-283 correlated with an improved GBM patient survival in clinical datasets and none of the proteins whose abundance was increased by TAT-Cx43266-283 correlated with shorter survival, supporting its use in clinical trials.

Finding the limits of deep learning clinical sensitivity with fractional anisotropy (FA) microstructure maps.

Quantitative maps obtained with diffusion weighted (DW) imaging, such as fractional anisotropy (FA) -calculated by fitting the diffusion tensor (DT) model to the data,-are very useful to study neurological diseases. To fit this map accurately, acquisition times of the order of several minutes are needed because many noncollinear DW volumes must be acquired to reduce directional biases. Deep learning (DL) can be used to reduce acquisition times by reducing the number of DW volumes. We already developed a DL network named "one-minute FA," which uses 10 DW volumes to obtain FA maps, maintaining the same characteristics and clinical sensitivity of the FA maps calculated with the standard method using more volumes. Recent publications have indicated that it is possible to train DL networks and obtain FA maps even with 4 DW input volumes, far less than the minimum number of directions for the mathematical estimation of the DT. Here we investigated the impact of reducing the number of DW input volumes to 4 or 7, and evaluated the performance and clinical sensitivity of the corresponding DL networks trained to calculate FA, while comparing results also with those using our one-minute FA. Each network training was performed on the human connectome project open-access dataset that has a high resolution and many DW volumes, used to fit a ground truth FA. To evaluate the generalizability of each network, they were tested on two external clinical datasets, not seen during training, and acquired on different scanners with different protocols, as previously done. Using 4 or 7 DW volumes, it was possible to train DL networks to obtain FA maps with the same range of values as ground truth - map, only when using HCP test data; pathological sensitivity was lost when tested using the external clinical datasets: indeed in both cases, no consistent differences were found between patient groups. On the contrary, our "one-minute FA" did not suffer from the same problem. When developing DL networks for reduced acquisition times, the ability to generalize and to generate quantitative biomarkers that provide clinical sensitivity must be addressed.

Unlocking the potential: T1-weighed MRI as a powerful predictor of levodopa response in Parkinson’s disease

BackgroundThe efficacy of levodopa, the most crucial metric for Parkinson’s disease diagnosis and treatment, is traditionally gauged through the levodopa challenge test, which lacks a predictive model. This study aims to probe the predictive power of T1-weighted MRI, the most accessible modality for levodopa response.MethodsThis retrospective study used two datasets: from the Parkinson’s Progression Markers Initiative (219 records) and the external clinical dataset from Ruijin Hospital (217 records). A novel feature extraction method using MedicalNet, a pre-trained deep learning network, along with three previous approaches was applied. Three machine learning models were trained and tested on the PPMI dataset and included clinical features, imaging features, and their union set, using the area under the curve (AUC) as the metric. The most significant brain regions were visualized. The external clinical dataset was further evaluated using trained models. A paired one-tailed t-test was performed between the two sets; statistical significance was set at p < 0.001.ResultsFor 46 test set records (mean age, 62 ± 9 years, 28 men), MedicalNet-extracted features demonstrated a consistent improvement in all three machine learning models (SVM 0.83 ± 0.01 versus 0.73 ± 0.01, XgBoost 0.80 ± 0.04 versus 0.74 ± 0.02, MLP 0.80 ± 0.03 versus 0.70 ± 0.07, p < 0.001). Both feature sets were validated on the clinical dataset using SVM, where MedicalNet features alone achieved an AUC of 0.64 ± 0.03. Key responsible brain regions were visualized.ConclusionThe T1-weighed MRI features were more robust and generalizable than the clinical features in prediction; their combination provided the best results. T1-weighed MRI provided insights on specific regions responsible for levodopa response prediction.Critical relevance statementThis study demonstrated that T1w MRI features extracted by a deep learning model have the potential to predict the levodopa response of PD patients and are more robust than widely used clinical information, which might help in determining treatment strategy.Key PointsThis study investigated the predictive value of T1w features for levodopa response.MedicalNet extractor outperformed all other previously published methods with key region visualization.T1w features are more effective than clinical information in levodopa response prediction.Graphical

Anogenital HPV-Related Cancers in Women: Investigating Trends and Sociodemographic Risk Factors

The incidences of anogenital HPV-related cancers in women are on the rise; this is especially true for anal cancer. Medical societies are now beginning to recommend anal cancer screening in certain high-risk populations, including high-risk women with a history of genital dysplasia. The aim of this study is to investigate national anogenital HPV cancer trends as well as the role of demographics, deprivation, and ethnicity on anogenital cancer incidence in England, in an attempt to better understand this cohort of women which is increasingly affected by anogenital HPV-related disease. Demographic data from the Clinical Outcomes and Services Dataset (COSD) were extracted for all patients diagnosed with anal, cervical, vulval and vaginal cancer in England between 2014 and 2020. Outcomes included age, ethnicity, deprivation status and staging. An age over 55 years, non-white ethnicity and high deprivation are significant risk factors for late cancer staging, as per logistic regression. In 2019, the incidences of anal and vulval cancer in white women aged 55–74 years surpassed that of cervical cancer. More needs to be done to educate women on HPV-related disease and their lifetime risk of these conditions.

A Clinical Bacterial Dataset for Deep Learning in Microbiological Rapid On-Site Evaluation

Microbiological Rapid On-Site Evaluation (M-ROSE) is based on smear staining and microscopic observation, providing critical references for the diagnosis and treatment of pulmonary infectious disease. Automatic identification of pathogens is the key to improving the quality and speed of M-ROSE. Recent advancements in deep learning have yielded numerous identification algorithms and datasets. However, most studies focus on artificially cultured bacteria and lack clinical data and algorithms. Therefore, we collected Gram-stained bacteria images from lower respiratory tract specimens of patients with lung infections in Chinese PLA General Hospital obtained by M-ROSE from 2018 to 2022 and desensitized images to produce 1705 images (4,912 × 3,684 pixels). A total of 4,833 cocci and 6,991 bacilli were manually labelled and differentiated into negative and positive. In addition, we applied the detection and segmentation networks for benchmark testing. Data and benchmark algorithms we provided that may benefit the study of automated bacterial identification in clinical specimens.

Exploring the potential of representation and transfer learning for anatomical neuroimaging: Application to psychiatry

The perspective of personalized medicine for brain disorders requires efficient learning models for anatomical neuroimaging-based prediction of clinical conditions. There is now a consensus on the benefit of deep learning (DL) in addressing many medical imaging tasks, such as image segmentation. However, for single-subject prediction problems, recent studies yielded contradictory results when comparing DL with Standard Machine Learning (SML) on top of classical feature extraction. Most existing comparative studies were limited in predicting phenotypes of little clinical interest, such as sex and age, and using a single dataset. Moreover, they conducted a limited analysis of the employed image pre-processing and feature selection strategies. This paper extensively compares DL and SML prediction capacity on five multi-site problems, including three increasingly complex clinical applications in psychiatry namely schizophrenia, bipolar disorder, and Autism Spectrum Disorder (ASD) diagnosis. To compensate for the relative scarcity of neuroimaging data on these clinical datasets, we also evaluate three pre-training strategies for transfer learning from brain imaging of the general healthy population: self-supervised learning, generative modeling and supervised learning with age. Overall, we find similar performance between randomly initialized DL and SML for the three clinical tasks and a similar scaling trend for sex prediction. This was replicated on an external dataset. We also show highly correlated discriminative brain regions between DL and linear ML models in all problems. Nonetheless, we demonstrate that self-supervised pre-training on large-scale healthy population imaging datasets (N≈10k), along with Deep Ensemble, allows DL to learn robust and transferable representations to smaller-scale clinical datasets (N≤1k). It largely outperforms SML on 2 out of 3 clinical tasks both in internal and external test sets. These findings suggest that the improvement of DL over SML in anatomical neuroimaging mainly comes from its capacity to learn meaningful and useful abstract representations of the brain anatomy, and it sheds light on the potential of transfer learning for personalized medicine in psychiatry

Post-reduction observation and recurrence of pediatric intussusception in New Zealand: A national multicenter retrospective study

BackgroundIntussusception is the most common cause of preschool intestinal obstruction. International data suggest that following uncomplicated enema reduction, 4-hour observation with prompt return to oral diet is safe. We aimed to investigate the rate and timing of intussusception recurrence across Aotearoa New Zealand, compared to the cost of routine post-reduction observation. MethodsWe performed a national 15-year retrospective, multicenter cohort study of all children treated for intussusception between 01 Jan 2007 and 01 Jan 2022 across AoNZ. A robust clinical dataset was collected using a REDCap proforma. We also investigated the financial and time costs of current in-hospital observation practices. ResultsDuring this period, primary enema reduction without general anesthetic was successful in 339 cases, requiring a median of 2 attempts (1–8). The median age was 0.8 years; 70.8% were male, and 17.1% were Māori. The median length of hospital stay was 25.9 h, primarily comprised of post-reduction observation (21.6 h). Intussusception recurrence within 30 days occurred in 29 children (8.5%) at a median time of 24.7 h post-reduction. Of the 19 cases recurring before discharge, 12 (3.5%) recurred between 4 and 24 h. The median observation period for those with successful primary enema reduction was 21.6 h; if reduced to 4 h, savings of NZD 883,632 (USD 529,825) and 7342 ward hours were possible. ConclusionsNationally, the intussusception recurrence rate after uncomplicated enema reduction was 8.5%, but only 3.5% occurred between the 4–24 hour period post-reduction. Expediting discharge by avoiding routine overnight observation appears safe and economically viable.

Leveraging federated learning for boosting data privacy and performance in IVF embryo selection.

To study the effectiveness of federated learning in in vitro fertilization on embryo evaluation tasks. This is a retrospective cohort analysis. Two datasets were used in this study. The ploidy status dataset consisted of 10,065 embryo records, 3760 treatments, and 2479 infertile couples from 5 hospitals. The clinical pregnancy dataset consisted of 4495 embryo records, 4495 treatments, and 3704 infertile couples from 4 hospitals. Federated learning and the gradient boosting decision tree algorithm were utilized for modeling. On the ploidy status dataset, the areas under the receiver operating characteristic curves of our model trained with federated learning were 71.78%, 73.10%, 69.39%, 69.72%, and 73.46% for 5 hospitals respectively, showing an average increase of 2.5% compared to those of our model trained without federated learning. On the clinical pregnancy dataset, the areas under the receiver operating characteristic curves of our model trained with federated learning were 72.03%, 56.77%, 61.63%, and 58.58% for 4 hospitals respectively, showing an average increase of 3.08%. Federated learning can improve data privacy and data security and meanwhile improve the performance of embryo selection tasks by leveraging data from multiple sources. This study demonstrates the effectiveness of federated learning in embryo evaluation, and the results show the promise for future application.

Prediction of Visual Acuity in Pseudophakic Cataract Population Based on Residual Refraction

Purpose The purpose of the study was to design a simple, handy prediction for the effect of spherical and cylindrical refractive error on the visual acuity degradation at different distances and validate this model on a clinical dataset. Methods This study examined 70 eyes from 35 patients’ post-cataract surgery with aberration-free intraocular lenses. Biometric and corneal data were analysed, and subjective refraction and visual acuity were evaluated by two experienced optometrists. The study computed the spherical equivalent (SEQ), and defocus equivalent via vector addition (DEQ vec), as the sum of absolute values (DEQ abs). Predictive models were developed using univariate regression, with confidence intervals (BCa 95%) calculated through non-parametric bootstrapping (10,000 cycles). Results Various calculated equivalents included −0.44 D for spherical equivalent (SEQ), 0.70 D for defocus equivalent based on vector calculation (DEQ vec), and 0.89 D for defocus equivalent based on absolute values (DEQ abs). Uncorrected and corrected visual acuity averaged 0.07 logMAR and −0.04 logMAR, respectively. The absolute defocus equivalent (DEQ abs) exhibited the smallest confidence interval (BCa 95%) at 0.07. Conclusion The defocus equivalent based on the addition of absolute values (DEQ abs) emerged as the most practical predictor for the described applications. Notably, it offers the advantage of easy calculability through a simple equation: VA loss = DEQ abs ⋅ 0.23. In 95% of cases, this predicted loss would have an accuracy of ±0.03 lines.