The Sociodemographic Biases in Machine Learning Algorithms: A Biomedical Informatics Perspective

Life, Journal Year: 2024, Volume and Issue: 14(6), P. 652 - 652

Published: May 21, 2024

Artificial intelligence models represented in machine learning algorithms are promising tools for risk assessment used to guide clinical and other health care decisions. Machine algorithms, however, may house biases that propagate stereotypes, inequities, discrimination contribute socioeconomic disparities. The include those related some sociodemographic characteristics such as race, ethnicity, gender, age, insurance, status from the use of erroneous electronic record data. Additionally, there is concern training data algorithmic large language pose potential drawbacks. These affect lives livelihoods a significant percentage population United States globally. social economic consequences associated backlash cannot be underestimated. Here, we outline sociodemographic, data, undermine sound medical decision-making should addressed system. We present perspective overview these by historically marginalized communities, bias, biased evaluations, implicit selection/sampling biases, distributions, cultural insurance conformation information bias anchoring make recommendations improve model including de-biasing techniques counterfactual role-reversed sentences during knowledge distillation, fine-tuning, prefix attachment at time, toxicity classifiers, retrieval augmented generation modification mitigate moving forward.

Language: Английский

---

IoT Health Devices: Exploring Security Risks in the Connected Landscape

IoT, Journal Year: 2023, Volume and Issue: 4(2), P. 150 - 182

Published: May 25, 2023

The concept of the Internet Things (IoT) spans decades, and same can be said for its inclusion in healthcare. IoT is an attractive target medicine; it offers considerable potential expanding care. However, application healthcare fraught with array challenges, also, through it, numerous vulnerabilities that translate to wider attack surfaces deeper degrees damage possible both consumers their confidence within health systems, as a result patient-specific data being available access. Further, when devices (IoTHDs) are developed, diverse range attacks possible. To understand risks this new landscape, important architecture IoTHDs, operations, social dynamics may govern interactions. This paper aims document create map regarding lay groundwork better understanding security emerging IoTHD modalities multi-layer approach, suggest means improved governance interaction. We also discuss technological innovations expected set stage novel exploits leading into middle latter parts 21st century.

Language: Английский

---

Patient Victimhood and the Risks of Using Artificial Intelligence Technology in Healthcare

Victimology, Journal Year: 2024, Volume and Issue: 10(4), P. 492 - 502

Published: Feb. 28, 2024

Artificial intelligence technologies are of increasing interest in the field medicine and one key areas for digital transformation healthcare . According to a number experts, medical professionals technology developers, use devices equipped with artificial will raise high level, which lead improved clinical decision-making, high-quality analysis images, prediction control correctness prescribed treatment .However, failures associated systems can have serious consequences both outcomes patients These could undermine public confidence health care institutions general Given certain novelty technological solutions, data on efficacy safety products currently considered insufficient .This publication raises two important issues The first part study describes main physical, social mental characteristics (properties) that increase likelihood they be victimized event crime situation innovative services second identifies risks using AI greatest concern those exploiting

Language: Английский

---

Opportunities for Artificial Intelligence in Operational Medicine: Lessons from the United States Military

Bioengineering, Journal Year: 2025, Volume and Issue: 12(5), P. 519 - 519

Published: May 14, 2025

Conducted in challenging environments such as disaster or conflict areas, operational medicine presents unique challenges for the delivery of efficient and quality healthcare. It exposes first responders medical personnel to many unexpected health risks dangerous situations. To tackle these issues, artificial intelligence (AI) has been progressively incorporated into medicine, both on front lines also more recently support roles. The ability AI rapidly analyze high-dimensional data make inferences opened up a wide variety opportunities increased efficiency its early adopters, notably United States military, non-invasive imaging mental applications. This review discusses current state highlights broad array potential applications developed military.

Language: Английский

---

High-Fidelity Synthetic Data Applications for Data Augmentation

Artificial intelligence, Journal Year: 2024, Volume and Issue: unknown

Published: Jan. 12, 2024

The use of high-fidelity synthetic data for augmentation is an area growing interest in science. In this chapter, the concept introduced, and different types are discussed terms their utility or fidelity. Approaches to generation presented compared with computer modelling simulation approaches, highlighting unique benefits data. One main applications supporting training validation machine learning algorithms, where it can provide a virtually unlimited amount diverse high-quality improve accuracy robustness models. Furthermore, address missing biases due under-sampling using techniques such as BayesBoost, well boost sample sizes scenarios real based on small sample. Another important application generating virtual patient cohorts, digital twins, estimate counterfactuals silico trials, allowing better prediction treatment outcomes personalised medicine. chapter concludes by identifying areas further research field, including developing more efficient accurate methods exploring ethical implications

Language: Английский

---

Comparative Analysis of Life Plans of Inmates of Different Regimes in Places of Deprivation of Liberty

Victimology, Journal Year: 2024, Volume and Issue: 10(4), P. 525 - 537

Published: Feb. 28, 2024

The relevance of the research topic is due to importance studying and analyzing life plans prisoners in prison. purpose this study was a comparative analysis different regimes (strict, general, colony-settlement). conducted on basis correctional institutions Federal Penitentiary Service Russia Ryazan region . It attended by 118 male convicts serving sentences various (general, strict, penal colony). Methods. During study, following methods were used: observation, conversation, questioning, personal files, methods: “Self-assessment plans” A. V. Naprisa (modified version “Self-Assessment Orientation” test G. Deev); “Time Perspective Questionnaire” F. Zimbardo; Attitude Scale” J. Nuytten, Lensom; Questionnaire “Style Explanation Successes Failures” (STONE-B) T.O. Gordeeva, Ye. N. Osina, Yu. Shevyakhova; “Vitality test” S. Maddi (adapted D. Leontiev, I. Rasskazova); mathematical statistics. Results. has been established that colony settlements have most constructive For general strict requires development psychocorrectional program will allow them realize values their life, model positive for future, admit guilt crime committed become law-abiding citizens

Language: Английский

---

Deep skin diseases diagnostic system with Dual-channel Image and Extracted Text

Frontiers in Artificial Intelligence, Journal Year: 2023, Volume and Issue: 6

Published: Oct. 19, 2023

Background Due to the lower reliability of laboratory tests, skin diseases are more suitable for diagnosis with AI models. There limited dermatology diagnostic models combining images and text; few these Asian populations, cover most common types diseases. Methods Leveraging a dataset sourced from Asia comprising over 200,000 220,000 medical records, we explored deep learning-based system Dual-channel extracted text model DIET-AI diagnose 31 diseases, which covers majority From 1 September December 2021, prospectively collected 6,043 cases records 15 hospitals in seven provinces China. Then performance was compared that six doctors different seniorities clinical dataset. Results The average not less than all seniorities. By comparing area under curve, sensitivity, specificity, demonstrate is effective scenarios. In addition, affect physicians varying degrees. Conclusion This largest dermatological Chinese demographic. For first time, built image classification on non-cancer dermatitis both achieved comparable senior about It provides references exploring feasibility evaluation use afterward.

Language: Английский

---

Victimblaming in Bribery

Victimology, Journal Year: 2024, Volume and Issue: 10(4), P. 463 - 473

Published: Feb. 28, 2024

The article deals with victimblaming in bribery. author of the article, based on results sociological research and law enforcement practice, patterns social phenomenon «victim blaming», puts forward substantiates hypothesis that there is a problem A number features stand out: bribery as characterized by determining real victim; dominant plotting generalization events narrative an act two guilty; prevailing comprehensive assessment guilty part both bribe taker taker; paradoxical nature (consideration various forms one whole, parallel paramount attention to receiving bribe); negative public giver increases significantly integrated approach corresponding their actions form it whole; greatest directed towards bribe-taker

Language: Английский

---

Current status of artificial intelligence methods for skin cancer survival analysis: a scoping review

Frontiers in Medicine, Journal Year: 2024, Volume and Issue: 11

Published: April 22, 2024

Skin cancer mortality rates continue to rise, and survival analysis is increasingly needed understand who at risk what interventions improve outcomes. However, current statistical methods are limited by inability synthesize multiple data types, such as patient genetics, clinical history, demographics, pathology reveal significant multimodal relationships through predictive algorithms. Advances in computing power science enabled the rise of artificial intelligence (AI), which synthesizes vast amounts applies algorithms that enable personalized diagnostic approaches. Here, we analyze AI used skin analysis, focusing on supervised learning, unsupervised deep natural language processing. We illustrate strengths weaknesses these approaches with examples. Our PubMed search yielded 14 publications meeting inclusion criteria for this scoping review. Most focused melanoma, particularly histopathologic interpretation learning. Such concentration a single type amid increasing focus learning highlight growing areas innovation; however, it also demonstrates opportunity additional addresses other types cutaneous malignancies expands scope prognostication combine both genetic, histopathologic, data. Moreover, researchers may leverage enhanced benefit analyses. Expanding arena improved targeted treatments,

Language: Английский

---

Understanding Disparities in Post Hoc Machine Learning Explanation

2022 ACM Conference on Fairness, Accountability, and Transparency, Journal Year: 2024, Volume and Issue: 144, P. 2374 - 2388

Published: June 3, 2024

Previous work has highlighted that existing post-hoc explanation methods exhibit disparities in fidelity (across "race" and "gender" as sensitive attributes), while a large body of focuses on mitigating these issues at the metric level, role data generating process black box model relation to remains largely unexplored. Accordingly, through both simulations well experiments real-world dataset, we specifically assess challenges originate from properties data: limited sample size, covariate shift, concept omitted variable bias, based properties: inclusion attribute appropriate functional form. Through controlled simulation analyses, our study demonstrates increased omission covariates increase disparities, with effect pronounced higher for neural network models are better able capture underlying form comparison linear models. We also observe consistent findings regarding shift bias Adult income dataset. Overall, results indicate explanations can depend properties. Based this systematic investigation, provide recommendations design mitigate undesirable disparities.

Language: Английский

---