Torbjörn E. M. Nordling

Performing proper image contrast adjustment without information loss is an art. The default settings are often inappropriate for the image in question rendering a contrast adjustment depending on trial and error. We propose a simple method, rank-based transformation (RBT), for image contrast adjustment that requires no prior knowledge. This makes RBT an ideal first tool to apply for underexposed images. The RBT algorithm normalizes and equalizes all the intensity differences of the image over the full intensity range of the image data type, and thus assigning equal weight to all gradients. Even the state-of-the-art AI tool Cellpose visually benefits from RBT preprocessing. Our comparison of histogram normalization methods demonstrates the ability of RBT to bring out image features.

Background: As we emerge from the COVID-19 pandemic and transition to a post-pandemic era, it is crucial to reflect on our experiences and prepare for future pandemics. Here we evaluate the impact of different methods for calculating the vaccine efficacy of COVID-19 vaccines, which has not been done previously. Methods: We conducted a meta-analysis of 38 approved COVID-19 vaccines using data from phase III clinical trials between May 4, 2020, and June 10, 2022. We analyze vaccine efficacy against multiple SARS-CoV-2 variants including the original strain, Alpha, Beta, Delta, and Kappa using multiple endpoints. Clinical endpoints are categorized into a tree structure including asymptomatic infection, symptomatic infection, mild to critical illness, and death. We employ re-estimated vaccine efficacies, including relative risk and Poisson regression with robust error variance, for equitable cross-vaccine comparisons. Results: We re-estimated 63 vaccine efficacies, revealing a 3% to 6% difference in five efficacies compared to the original study. Four efficacies exhibited lower bounds below the critical 50% threshold for the endpoint asymptomatic, symptomatic, moderate, and severe, contrary to the initial reports. However, efficacy consistently surpasses the 50% threshold against symptomatic COVID-19. Overall efficacies range from 34.2% to 100%, 50.3% to 100% against symptomatic, and 66.8% to 100% against severe, and 65% to 95% against variants. Conclusions: Our systematic classification of vaccine endpoints enables more statistically rigorous meta-analyses across studies. Beyond the quantitative results, our study emphasizes the need to standardize the estimation method for robust assessments of vaccine efficacy. We highlight the incompleteness of the knowledge about different vaccine efficacy in the middle of the pandemic, in particular the need to identify variants during the trials and report on multiple endpoints. We encourage all authors to publicly share their data, fostering additional impartial investigations. This data collection enables comparisons with real-world effectiveness data, enabling future studies of the predictive power of efficacy.

Proper image contrast adjustment without information loss is essential but challenging, often requiring experience or trial and error. Nonetheless, we expected that the multiple normalization steps embedded in deep neural networks would render external contrast adjustment redundant in modern cell segmentation tools--but that is not the case. We evaluate the impact of contrast preprocessing on three state-of-the-art segmentation tools: Cellpose, PlantSeg, and Ilastik. Contrast adjustment via the rank-based transformation (RBT) method requires no parameter tuning or prior knowledge. RBT computes pixel ranks and maps them across the full dynamic range, ensuring uniform contrast enhancement without information loss. It is particularly effective for underexposed images. Preprocessing with RBT leads to visibly improved segmentation in advanced tools like Cellpose. Our results falsify the claim that modern segmentation tools are insensitive to contrast preprocessing.

This paper addresses the fundamental computer vision challenge of robust circle detection and fitting in degraded imaging conditions. We present Combinatorial Convolution-based Circle Fitting for Blurry Images (3C-FBI), an algorithm that bridges the gap between circle detection and precise parametric fitting by combining (1) efficient combinatorial edge pixel (edgel) sampling and (2) convolution-based density estimation in parameter space. We evaluate 3C-FBI across three experimental frameworks: (1) real-world medical data from Parkinson's disease assessments (144 frames from 36 videos), (2) controlled synthetic data following established circle-fitting benchmarks, and (3) systematic analysis across varying spatial resolutions and outlier contamination levels. Results show that 3C-FBI achieves state-of-the-art accuracy (Jaccard index 0.896) while maintaining real-time performance (40.3 fps), significantly outperforming classical methods like RCD (6.8 fps) on a standard CPU (i7-10875H). It maintains near-perfect accuracy (Jaccard almost 1.0) at high resolutions (480x480) and reliable performance (Jaccard higher than 0.95) down to 160x160 with up to 20% outliers. In extensive synthetic testing, 3C-FBI achieves a mean Jaccard Index of 0.989 across contamination levels, comparable to modern methods like Qi et al. (2024, 0.991), and surpassing RHT (0.964). This combination of accuracy, speed, and robustness makes 3C-FBI ideal for medical imaging, robotics, and industrial inspection under challenging conditions.

The COVID-19 pandemic has demonstrated the shortcomings of epidemiological modelling for guiding policy decisions. Moreover, the modelling efforts resulted in many models yielding different predictions, creating a need to compare these predictions to determine which model is most accurate. We introduce a data synthesis algorithm, CovSyn, designed to generate synthetic COVID-19 datasets providing sufficiently detailed information for benchmarking epidemiological models against a known synthetic ground truth. CovSyn utilizes observed infections with contact tracing, testing, course of disease data, and a contact network based on municipality statistics, which categorises connections into household, school, workplace, healthcare, and municipality. The model’s initial parameters and boundaries are derived from empirical data, including the first community outbreak of COVID-19 in Taiwan and clinical observations. Comprehensive parameter space exploration for optimal results is done by the Firefly algorithm. We demonstrate it and validate our estimates by comparing state transition times, daily social contacts, and associated secondary attack rates against a structured dataset and clinical observations. Our simulations align with prior research and this dataset. Most state transition times from 10,000 simulations are within uncertainty ranges. Daily contact numbers and their distribution across layers match empirical findings. Our model accurately reproduced the first COVID-19 outbreak in Taiwan, achieving high accuracy with observed cumulative confirmed cases ( R ² = 0.9) across daily, 7-day moving average, and 31-day moving average levels. Each synthetic subject contains demographic data (age, gender, occupation), course of disease (latent/incubation periods, testing, isolation, critical illness, recovery, and death dates), and contact network data including daily interactions with infected and uninfected individuals. Our algorithm offers a valid alternative for developing and benchmarking epidemiological models to advance COVID-19 forecasting research. Author summary: We present a novel approach to enhance the testing of epidemiological infection modelling and demonstrate spread prediction accuracy by testing COVID-19 simulation models on Taiwanese synthetic data with known ground truth. CovSyn generates comprehensive synthetic data at the individual level, encompassing demographic characteristics (age, gender, occupation), course of disease (infection dates, symptom onset, recovery dates), and contact tracing information (daily social interactions including household, school, workplace, healthcare, and municipality). It can provide a consistent, standardized synthetic dataset for model evaluation, addressing the previous challenge of comparing COVID-19 models that used disparate data sources and different time periods. In this study, we detail the algorithm and demonstrate its reliability by creating a synthetic dataset for the first outbreak of SARS-CoV-2 in Taiwan and comparing it with the collected Taiwan COVID-19 dataset. Future work will focus on benchmarking state-of-the-art forecasting models using our synthetic data. Through CovSyn’s detailed individual-level data, we aim to advance the development of more accurate epidemiological models.

The COVID-19 pandemic has flooded open databases with population-level data. However, individual-level structured data, such as the course of disease and contact tracing information, is almost non-existent in open databases. Publish a structured and cleaned COVID-19 dataset with the course of disease and contact tracing information for easy benchmarking of COVID-19 models. We gathered data from Taiwanese open databases and daily news reports. The outcome is a structured quantitative dataset encompassing the course of the disease of Taiwanese individuals, alongside their contact tracing information. Our dataset comprises 579 confirmed cases covering the period from January 21, to November 9, 2020, when the original SARS-CoV-2 virus was most prevalent in Taiwan. The data include features such as travel history, age, gender, symptoms, contact types between cases, date of symptoms onset, confirmed, critically ill, recovered, and dead. We also include the daily summary data at population-level from January 21, 2020, to May 23, 2022. Our data can help enhance epidemiological modelling.