This study examines how machine learning (ML) systems—often used to support healthcare decisions—can sometimes produce unfair results, meaning their predictions may be less accurate for certain patient groups. A key concern is that clinicians may rely too heavily on these systems, which can unintentionally reinforce these biases. The researchers explored a strategy called��ML suppression, which means selectively “silencing” or withholding certain AI predictions when they are likely to be unreliable, based on an auditing process. They also looked at whether incorporating��uncertainty estimates��(how confident the model is in its predictions) could help decide when to suppress outputs.

Using large hospital datasets, the team simulated how clinicians and ML systems would work together to predict outcomes like death, ICU admission, or hospital readmission. They compared different scenarios, including when the AI performed better than clinicians and when it performed worse. They evaluated both accuracy (using a standard metric called AUC, which measures how well predictions distinguish outcomes) and fairness (measured by differences in error rates across groups).

The results showed that when the AI model performed better than clinicians, using suppression improved overall performance without making fairness worse. When clinicians performed better, relying on human judgment alone was often as fair or fairer than using suppressed AI predictions. Importantly, adding uncertainty information helped improve results further by better identifying when AI predictions should be ignored. Overall, the study suggests that carefully filtering out low-quality AI predictions can improve both the effectiveness and fairness of human–AI collaboration in healthcare.

��

Figure 1.

Schematic indicating the collaboration scenario with and without suppression.³⁴

This article examines the experience of the NIH’s Bridge2AI Program, which funded four large biomedical and behavioral datasets designed to be well documented and ready for use with artificial intelligence (AI) and machine learning (ML). The goal of these datasets is to encourage responsible and effective use of AI in research, but building them raised many ethical, legal, social, and practical challenges. The authors describe the key steps involved in creating and managing these AI-ready datasets, including deciding which data to collect and why, responding to public concerns, handling participant consent based on how the data were obtained, ensuring responsible future use, determining where and how data are stored, clarifying how much control participants have over data sharing, and setting rules for data access and downloading.

Across these steps, the projects faced important questions about long-term data storage, future uses of the data, and how to balance openness with privacy and participant protection. The authors highlight the different choices made by the four projects, such as how they gathered public input, selected data storage solutions, and defined criteria for who can access and download the data. Although the governance approaches varied, common themes emerged, suggesting shared best practices.

Overall, the article summarizes key lessons learned from the Bridge2AI Program about how to collect, manage, and govern large datasets intended for AI and ML. These insights can guide future initiatives in designing datasets that are not only technically useful for AI, but also ethically sound, socially responsible, and trustworthy.

Figure 1.

Steps in governance of data collection and decision-making and responsible use for the development of AI with greater attention to public concerns throughout. The first 2 steps—promoting responsible selection—address the primary work of the DGPs, while the remaining 4 steps—promoting responsible use—are crucial factors the DGPs must consider.

]]> /valiant/2025/11/23/harmfulness-score-a-data-driven-framework-for-ranking-environmental-risks-of-microplastics/ Sun, 23 Nov 2025 16:55:16 +0000 /valiant/?p=5484 Souza, Fernando Gomes., Bhansali, Shekhar., & Thundat, Thomas G. (2025).��.��Macromolecular Rapid Communications. Advance online publication.��

In this study, researchers analyzed��104,471 scientific abstracts��about��microplastics and nanoplastics��using both��bibliometric tools��(methods for studying patterns in scientific publications) and��machine learning models. This allowed them to map out major research themes and identify which plastic materials are most often linked to potential health and environmental risks.

They created a combined��Harmfulness Score��by using��sentiment analysis, descriptions of harmful effects, and measures of how central certain terms are in scientific networks. Using this score,��polystyrene (PS)��and��polyethylene (PE)��ranked as the plastics most frequently associated with terms like��oxidative stress,��cytotoxicity, and��genotoxicity, indicating that they are discussed more often in connection with harmful biological effects.

The analysis also revealed that important��physicochemical details—such as particle size, density, and surface area—were rarely reported across studies (only 3.91%, 0.01%, and less than 0.01% of abstracts, respectively). This lack of information makes it harder to build accurate computer models or conduct reliable risk assessments.

Thematic clustering showed that current research focuses heavily on��environmental policy��and��biological impacts��of microplastics, while newer areas such as��microbial and enzymatic degradation��and��legal-policy connections��are rapidly emerging.

Overall, the findings point to a clear need for more��standardized reporting practices��and broader use of consistent analytical frameworks to improve the reliability of research and support better policymaking.

FIGURE 1

Co-occurrence network of terms in microplastic and nanoplastic research from 1961 to 2025, generated using VOSviewer. (a) Network visualization map showing clustered thematic areas based on term co-occurrence frequency and total link strength. (b) Overlay visualization map displaying the average publication year of terms, highlighting temporal trends and emerging research topics.

]]> /valiant/2025/09/26/modeling-the-mri-gradient-system-with-a-temporal-convolutional-network-improved-reconstruction-by-prediction-of-readout-gradient-errors/ Fri, 26 Sep 2025 19:50:23 +0000 /valiant/?p=5180 Martin, Jonathan B., Alderson, Hannah E., Gore, John C., Does, Mark D., & Harkins, Kevin D. (2025). Magnetic Resonance in Medicine.

The goal of this study was to develop a model that can better predict and correct distortions in MRI images caused by imperfections in the gradient system, which is responsible for shaping the magnetic fields used to create images. To do this, we trained a temporal convolutional network—a type of machine learning model—using data collected from a small animal imaging system. This network learned to predict the actual gradient waveforms produced by the scanner, including the nonlinear distortions that often reduce image quality.

When we incorporated these predictions into the image reconstruction process, the results showed clearer images and more accurate mapping of diffusion parameters compared to standard methods. This means that our approach outperformed both the use of the system’s default gradient settings and the commonly used gradient impulse response function. Overall, this work shows that temporal convolutional networks provide a more accurate way to model gradient behavior and can be used to correct errors after scanning, ultimately improving the quality and reliability of MRI images.

FIGURE 1

Spiral and chirp gradient waveforms measured on the 7T Bruker system, with conspicuous nonlinearities. Waveform timecourses are normalized by their respective maximum nominal amplitudes. (A,B) show the nominal waveforms with zoomed-in ROI highlighted. (C,D) show the nominal and measured waveforms at several different amplitudes. Clear nonlinearities are present. In (C) the two waveforms are not simply scaled copies of one another but have distinct zero crossing artifacts. In (D), the response varies with amplitude of the applied waveform. Delay and attenuation increase with decreasing amplitude

]]> /valiant/2025/08/25/what-are-the-future-directions-for-microplastics-characterization-a-regex-llama-data-mining-approach-for-identifying-emerging-trends/ Mon, 25 Aug 2025 21:03:50 +0000 /valiant/?p=5043 Gomes, Fernando, Bhansali, Shekhar, da Silveira Maranhão, Fabiola, Valladão, Viviane Silva, & Velasco, Karine. (2025). “.” Anais da Academia Brasileira de Ciências, 97, e20241345.

This study presents a new hybrid method to identify and analyze techniques used to study microplastics. By combining pattern-recognition software (regex) with the Llama 3.2:3b language model, we can better detect and understand both traditional and emerging techniques. Established methods like Raman and FTIR spectroscopy are examined alongside advanced tools such as X-ray Photoelectron Spectroscopy (XPS) and Surface-Enhanced Raman Spectroscopy (SERS). This approach improves both the speed and accuracy of identifying complex terms used in microplastics research. Using VOSDataAnalyzer and VOSviewer, we mapped connections and trends among related terms, identifying the 15 most commonly used and emerging techniques. Our analysis shows a shift toward more sensitive and innovative methods in microplastic studies. This Regex-Llama approach, introduced here for the first time, can be applied broadly to tasks such as studying pollutants in the environment, evaluating material breakdown in engineering, and assessing the health impacts of tiny contaminants. Overall, this strategy helps support environmental assessments and guide pollution reduction efforts across multiple fields.

Figure 1. �� Representation of the chemical structure of the most common polymers found in microplastic pollution, in sequence: Polyethylene (PE), Polypropylene (PP), Polystyrene (PS), and Polyethylene Terephthalate (PET).

]]> /valiant/2025/08/25/role-and-use-of-race-in-artificial-intelligence-and-machine-learning-models-related-to-health/ Mon, 25 Aug 2025 20:46:04 +0000 /valiant/?p=5032 Were, Martin Chieng, Li, Ang, Malin, Bradley A., Yin, Zhijun, Coco, Joseph R., Collins, Benjamin Xavier, Clayton, Ellen Wright, Novak, Laurie Lovett, Hendricks-Sturrup, Rachele M., & Oluyomi, Abiodun Olufemi. (2025). “.” Journal of Medical Internet Research, 27, e73996.

The use of race in health-related artificial intelligence (AI) and machine learning (ML) models has become a topic of growing attention and debate. Despite the many complex issues involved, there is currently no clear framework to help guide researchers, developers, and other stakeholders in examining and addressing these challenges. This perspective offers a broad, organized overview of the problems related to race in AI and ML, structured around the typical steps in developing and using these models. It also provides “points to consider” to help guide thoughtful inquiry and decision-making.

Figure 1. An artificial intelligence and machine learning life cycle model used to frame discussion on race. Adapted from Collins et al [17]. AI: artificial intelligence; ML: machine learning.

Malalignment, or incorrect positioning, is a major cause of implant failure after total knee replacement surgery (TKA). Checking alignment by manually analyzing medical images for many patients is not practical, so machine learning (ML) models could help by quickly and accurately measuring alignment and identifying patients at risk of problems. This study aimed to develop an ML model that can accurately determine knee alignment from full-length X-ray images showing the leg from hip to ankle.

The researchers collected long-leg X-rays from 550 patients who had knee replacement surgery. They used 440 of these images to train the ML model to identify key landmarks on the bones and implants, such as the hip joint, parts of the thigh and shin bones, and the implanted knee components. Using these landmarks, the model calculated important alignment angles used by doctors to evaluate knee positioning. The remaining 110 X-rays were used to test how accurate the model was.

The ML model was very fast, analyzing each image in less than 0.1 seconds. It measured alignment angles with very small errors compared to human measurements—less than one degree difference on average for all the angles tested.

In conclusion, this ML model shows high accuracy in assessing knee alignment after surgery and demonstrates great potential to improve clinical workflows and boost research in joint replacement care.

Figure 2��

Image augmentation process. From left to right, original image, vertically flipped image, horizontally flipped image, random rotated image, resized without padding image, and random brightness contrast changed image.

��

��

Combining traditional chemical information (called “descriptors”) with advanced machine learning models known as graph neural networks (GNNs) offers a promising way to improve how scientists screen potential drug molecules—a process known as ligand-based virtual screening. In this study, researchers tested how well different types of GNNs benefited from this combination.��

They found that some models, like GCN and SchNet, improved a lot when descriptors were added, while another model, SphereNet, only improved slightly. Still, all three models—GCN, SchNet, and SphereNet—performed about the same when this combination approach was used. This suggests that even simpler GNNs can work just as well as more complex ones if you add the right chemical information.��

The researchers also discovered that a set of expert-designed descriptors performed very well on their own, especially in challenging tests that mimic real-world drug discovery. In fact, these descriptors sometimes did better than the GNN-descriptor combinations. This highlights the need for new GNN models that can handle these real-world challenges more effectively.��

Overall, this study shows that blending chemical knowledge with machine learning can improve virtual drug screening and points to new directions for improving these tools in the future. The code for this work is available at��.��

��

Figure 1. Overview of the investigated method. The learned molecular representation of GNN is concatenated with expert-crafted descriptors to enhance the predictive power.��

��

Contrast-enhanced CT scans��use special dyes (called radiocontrast agents) that highlight blood vessels by making them appear brighter than the surrounding tissue. To get the best images, it’s important to time the scan correctly—right when the contrast is at its strongest in the area being examined.��

This study introduces a new method for predicting the��optimal timing of contrast��during a CT scan using a type of machine learning called a��circular regression model. Instead of treating contrast timing as one of a few fixed stages (as many previous methods do), this technique treats timing as a��continuous��value. That allows for more precise predictions and better adjustment to differences between patients—especially how contrast flows through each person’s blood vessels.��

The model uses��2D convolutional neural networks��(a kind of AI that processes image data) to learn patterns from earlier time points in a scan and predict the best contrast timing. It was trained on 877 CT scans and tested on 112 new scans, achieving��93.8% accuracy, which is on par with the best current methods. The results show that this new approach, which focuses on prediction rather than classification, performs better than existing 2D and 3D models that try to label scans into fixed categories.��

The study also investigates how the��location of each CT slice in the body��relates to contrast timing, suggesting that this information could help make predictions even more accurate—a new idea that hasn’t been explored before.��

��

Fig. 1.����

Four contrast phases used as anchor points in our re-gression model. Organs (e.g. kidneys) are enhanced differently in each phase. The difference between some phases are subtle, e.g., between Nand D, making automatic identification challenging.��

��

“Fingerprinting” is a method used to identify devices based on their unique data patterns—kind of like how human fingerprints are used to tell people apart. This paper focuses on��sensor fingerprinting, which means identifying individual sensors by the tiny, unique errors they have due to small imperfections from the manufacturing process.

The researchers created a model that uses these small error patterns to recognize specific sensors. They tested it using different��machine learning algorithms, including a��random forest classifier,��multilayer perceptron, and��soft decision tree. These techniques were able to correctly identify sensors with high accuracy—87%, 85.5%, and 89.2%, respectively.

These findings show that sensor fingerprinting is a promising and reliable way to identify or track sensors, which could be useful in areas like security, quality control, and device management.��

Fig. 1:����

High-level architecture of the testbed.��

��