Natural language processing for chest X‐ray reports in the transformer era: BERT‐like encoders for comprehension and GPT‐like decoders for generation

iRadiology, Journal Year: 2025, Volume and Issue: unknown

Published: Jan. 6, 2025

We conducted a comprehensive literature search in PubMed to illustrate the current landscape of transformer-based tools from perspective transformer's two integral components: encoder exemplified by BERT and decoder characterized GPT. Also, we discussed adoption barriers potential solutions terms computational burdens, interpretability concerns, ethical issues, hallucination problems, malpractice, legal liabilities. hope that this commentary will serve as foundational introduction for radiologists seeking explore evolving technical chest X-ray report analysis transformer era. Natural language processing (NLP) has gained widespread use computer-assisted (CXR) analysis, particularly since renaissance deep learning (DL) 2012 ImageNet challenge. While early endeavors predominantly employed recurrent neural networks (RNN) convolutional (CNN) [1], revolution is brought [2] its success can be attributed three key factors [3]. First, self-attention mechanism enables simultaneous multiple parts an input sequence, offering significantly greater efficiency compared earlier models such RNN [4]. Second, architecture exhibits exceptional scalability, supporting with over 100 billion parameters capture intricate linguistic relationships human [5]. Third, availability vast internet-based corpus advances power have made pre-training fine-tuning large-scale feasible [6]. The development resolution previously intractable problems achieves expert-level performance across broad range CXR analytical tasks, name entity recognition, question answering, extractive summarization [7]. In commentary, (Figure 1) landscape, barriers, handling comprehension managing generation. As our primary focus NLP, classification criteria or was based on text modules excluded research purely focusing vision transformers (ViT). Literature pipeline identify relevant articles published June 12, 2017, when model first introduced, October 4, 2024. followed previous systematic reviews [3, 8, 9] design groups keywords: (1) "transformer"; (2) "clinical notes", reports", narratives", text", "medical text"; (3) "natural processing", "text mining", "information extraction"; (4) "radiography", "chest film", radiograph", "radiograph", "X-rays". means communication between referring physicians, reports contain high-density information patients' conditions [10]. Much like physicians interpreting reports, step NLP understanding content important application explicitly converting it into format suitable subsequent tasks. One notable [11], which stands bidirectional representations transformers. contrast predecessors rely large amounts expert annotations supervised [12], undergoes self-supervised training unlabeled datasets understand patterns subsequently fine-tuned small set target task [12, 13], yielding superior [14], recognition [15], [16], semantics optimization [17]. context healthcare, Olthof et al. [18] built evaluate varying complexities, disease prevalence, sample sizes, demonstrating statistically outperformed conventional DL CNN, area under curve F1-score, t-test p-values less than 0.05. Beyond models, adapting domain-specific further enhance effectiveness various Yan [19] adapted four BERT-like encoders using millions radiology tackle tasks: identifying sentences describe abnormal findings, assigning diagnostic codes, extracting summarize reports. Their results demonstrated adaptation yielded significant improvements accuracy, ROUGE metrics all Most BERT-relevant studies sentence-, paragraph-, report-level predictions, while are also well-suited word-level pattern recognition. Chambon [20] leveraged [21], biomedical-specific BERT, probability individual tokens containing protected health information, replaced identified sensitive synthetic surrogates ensure privacy preservation. Similarly, Weng [22] developed system utilizing ALBERT [23], lite reduced parameters, keywords unrelated thereby reducing false-positive alarms outperforming regular expression-, syntactic grammar-, DL-based baselines. BERT-derived labels applied develop targeting other modalities 13]. Nowak [24] systematically explored utility BERT-generated silver linked them corresponding radiographs image classifiers. Compared trained exclusively radiologist-annotated gold labels, integrating led improved discriminability. macro-averaged synchronous proved effective settings limited whereas silver, better cases abundant labels. Zhang [25] introduced novel approach more generalizable classifiers, rather relying predefined categories: first, they used extract entities relationships; second, constructed knowledge graph these extractions; third, refined their domain expertise. Unlike traditional multiclass established not only categorized each but revealed interpretable categories, those linking anatomical regions signs. addition deriving advanced capabilities unprecedented innovation: direct supervision pixel-level segmentation medical [26]. Li [26] proposed text-augmented lesion paradigm integrated BERT-based textual compensate deficiency radiograph quality refine pseudo semi-supervision. These highlight strength comprehending healthcare-related annotation systems multi-modality beyond text. Meanwhile, researchers failures complex clinical Sushil [27] implementations inference achieved test accuracy 0.778. adaptations textbooks 0.833, still fell short experts. Potential limitations lie relatively modest parameter size, although larger reliance inadequate corpora, books, Wikipedia, selected databases [28]. Consequently, ability learn remains constrained. shortcomings being alleviated GPT-like decoders, incorporate hundreds billions internet-scale corpora [29]. Following advent encoders, generative pre-trained (GPT) [30], next groundbreaking leap, breaks enabling non-experts perform tasks through freely conversational without any coding. CvT2DistilGPT2 [31], prominent generator era, utilizes ViT GPT-2 decoder. experiments indicated CNN GPT surpassed encoder–decoder architectures specific generation applications, state-of-the-art methods integrate decoders. TranSQ [32] framework. emulates reasoning process generating reports: formulating hypothesis embeddings represent implicit intentions, querying visual features extracted synthesizing semantic cross-modality fusion, transforming candidate DistilGPT [33]. Finally, attained BLEU-4 score 0.205 0.409. comparison, best-performing baseline among 17 retrieval 0.188 0.383, highlighting capability unified multi-modality. Though decoders dominated general domain, family long short-term memory (LSTM) [34] good partially because highly templated characteristics [32]. Kaur Mittal [35] classical architectures, feature extraction, LSTM token They modules, generate numerical inputs prior shortlist disease-relevant afterward. Results presented solution 0.767 0.897, suggesting approaches remain viable backbone scenarios. quantitative comparing outputs ground truth model-generated should supplemented evaluation Boag [36] study automated generation, divergence accuracy. A discrepancy readability been reported [37]. Accordingly, emphasize involvement rating correctness readability. sections, reviewed applications Although remarkable well-established, face problems. Some integration specialized expertise [31, 38], others necessitate resolution. demands era substantial. For example, version contains 334 million GPT-3 175 billion. contrast, support vector machines [39] random forests [40], require few hundred thousand parameters. result, many healthcare providers cannot afford costs tailoring scratch. To address this, offer several recommendations. development, suggest leverage open-access building fine-tuning, considering scales, recommend parameter-efficient technique updates subset model's leaving majority weights unchanged [41]. An exemplificative Taylor [42] empirically validated techniques within domain. advocate prompt engineering techniques, retrieval-augmented crafting informative instructive guide decoders' output changing [43]. Ranjit [44] method retrieve most contextual prompts concise accurate retaining critical entities. Last least, obtaining approval ethics committees share anonymous data facilitate collaboration external partners, helping alleviate resource burdens. including both where decisions directly impact lives. often regarded black-box simple render explainable modern layers neurons dissected visualized, providing insights functionality [45-48]. behavior challenge due complexity associated exponential scaling neuron numbers [49]. though internal activations challenging interpret, preliminary analyzing influence high degree alignment assessments [50, 51]. lies flexibility align instructions. This allows users obtain expected request explanations outputs, fostering enhanced usability [52, 53]. readers overview detailed insights, surveys [54-56]. considerations paramount transformers, given powerful nuanced datasets. concerns pressing private representative population. patient privacy, anonymizing during deployment stages neither learned [57] nor inadvertently disclosed certain [58]. Dataset representativeness issue, underrepresentation minority exacerbate disparities perpetuate inequities [59]. mitigate risk, developers prioritize inclusivity collection, maintainers continuously monitor equitable outcomes [60]. Fourth, coherent responses diverse user solving wide [61], predictive internet instead radiological well-defined logic [62]. Therefore, continue suffer hallucinations, phenomenon appears plausible factually incorrect, nonsensical, users' [63]. Current efforts broadly post-training stages. During training, strategies include in-house reinforcement guided radiologists' feedback 64]. Post-training encompass detection, knowledge, multi-agent collaboration, radiologist-in-the-loop frameworks [62, 65]. Due space constraints, encourage refer 66-68] strategies. Lastly, even after refinements, may present risks potentially leading errors liabilities [69]. Errors arise sources, inaccurate clinician nonadherence correct recommendations, poor workflows [70]. determining responsibility adverse issue stakeholders, software developers, maintenance teams, departments, [71]. European Commission focuses safety liability implications artificial intelligence, applies device laws demonstrates generally falls civil product Civil typically pertains developers. However, stops strict definitive framework inherent ambiguity algorithms questions surrounding likely addressed courts case law. Under existing frameworks, follow standard care, supplementary confirmatory substitutes practice beneficial stakeholders Additionally, departments implement tools, involve radiologists, throughout entire cycle [72], prepare in-depth programs familiarize differ routine statistical tests black boxes resist full interpretation [73]. Moreover, expectations important: unrealistic optimism, seen replacement expertise, undue pessimism, perceived no utility, avoided [74-77]. Han Yuan: Conceptualization; curation; formal analysis; investigation; project administration; validation; visualization; writing—original draft; writing—review editing. None. author declares he conflicts interest. exempt review committee does participants, animal subjects, collection. Not applicable. Data sharing apply were generated analyzed.

Language: Английский

---

Representation of intensivists’ race/ethnicity, sex, and age by artificial intelligence: a cross-sectional study of two text-to-image models

Critical Care, Journal Year: 2024, Volume and Issue: 28(1)

Published: Nov. 11, 2024

Integrating artificial intelligence (AI) into intensive care practices can enhance patient by providing real-time predictions and aiding clinical decisions. However, biases in AI models undermine diversity, equity, inclusion (DEI) efforts, particularly visual representations of healthcare professionals. This work aims to examine the demographic representation two text-to-image models, Midjourney ChatGPT DALL-E 2, assess their accuracy depicting characteristics intensivists.

Language: Английский

---

Advancing Ethical Considerations for Data Science in Injury and Violence Prevention

Public Health Reports, Journal Year: 2025, Volume and Issue: unknown

Published: Jan. 20, 2025

Data science is an emerging field that provides new analytical methods. It incorporates novel data sources (eg, internet data) and methods machine learning) offer valuable timely insights into public health issues, including injury violence prevention. The objective of this research was to describe ethical considerations for scientists conducting prevention–related projects prevent unintended ethical, legal, social consequences, such as loss privacy or trust. We first reviewed foundational bioethics ethics literature identify key concepts relevant science. After identifying these concepts, we held a series discussions organize them under broad domains. Within each domain, examined from our review the primary literature. Lastly, developed questions domain facilitate early conceptualization stage analysis prevention projects. identified 4 domains: privacy, responsible stewardship, justice fairness, inclusivity engagement. determined carries equal weight, with no consideration bearing more importance than others. Examples are clearly project goals, determining whether people included in at risk reidentification through external linkages, evaluating minimizing potential bias used. As methodologies incorporated work toward reducing effect on individuals, families, communities United States, recommend issues be identified, considered, addressed.

Language: Английский

---

<b>Memanfaatkan Kecerdasan Buatan dan Pembelajaran Mesin dalam Inovasi Farmasi</b>

MEDICINUS, Journal Year: 2025, Volume and Issue: 38(2), P. 28 - 35

Published: Feb. 1, 2025

Integrasi kecerdasan buatan (artificial intelligence/AI) dan pembelajaran mesin (machine learning/ML) telah merevolusi industri farmasi, mengubah cara obat ditemukan, dikembangkan, diuji, diproduksi. Teknologi ini memungkinkan efisiensi akurasi yang belum pernah terjadi sebelumnya dengan memanfaatkan sejumlah besar data algoritmakomputasi canggih. Dalam penemuan obat, AI mempercepat identifikasi target terapeutik desain molekul baru, secara drastis mengurangi waktu menuju pemasaran. Selama pengembangan, ML membantu mengoptimalkan uji klinik stratifikasi populasi pasien untuk meningkatkan presisi efektivitas. klinik, alat berbasis rekrutmen, pemantauan, adaptif, menghasilkan studi lebih andal hemat biaya. Terakhir, memastikan pengendalian kualitas real-time pemeliharaan prediktif dalam manufaktur, konsistensi produk biaya operasional. Makalah mengeksplorasi aplikasi AI/ML komprehensif di berbagai domain, didukung oleh kasus analisis mendalam tentang dampaknya. Selain itu, makalah membahas tantangan seperti data, hambatan regulasi, transparansi algoritma menghambat adopsinya luas. Pertimbangan etis, termasuk masalah privasi risiko bias sistem juga dievaluasi. Akhirnya, menguraikan peluang kemajuan masa depan, menekankan perlunya upaya kolaboratif antara akademisi, industri, badan regulasi potensi penuh membentuk kembali lanskap farmasi.

---

Integrating enterprise risk management to address AI‐related risks in healthcare: Strategies for effective risk mitigation and implementation

Journal of Healthcare Risk Management, Journal Year: 2025, Volume and Issue: unknown

Published: Feb. 14, 2025

Abstract The incorporation of artificial intelligence (AI) in health care offers revolutionary enhancements patient diagnostics, clinical processes, and overall access to services. Nevertheless, this technological transition brings forth various new, intricate risks that pose challenges current safety ethical norms. This research explores the ability enterprise risk management as an all‐encompassing framework tackle these arising risks, providing both a forward‐looking responsive strategy designed for industry. At core method are instruments together seek proactively uncover address AI‐related weaknesses like algorithmic bias, system failures, data privacy issues. On reactive side, it incorporates incident reporting systems root cause analysis, tools enable providers quickly unexpected events consistently improve AI implementation procedures. However, some application difficulties still exist. unclear, “black box” characteristics numerous models hinder transparency responsibility, prompting inquiries about clarity AI‐generated choices their adherence benchmarks treatment. highlights with progress technologies, also needs evolve, addressing new complexities while promoting culture focused on settings.

Language: Английский

---

A Narrative Review in Application of Artificial Intelligence in Forensic Science: Enhancing Accuracy in Crime Scene Analysis and Evidence Interpretation

Journal of international oral health, Journal Year: 2025, Volume and Issue: 17(1), P. 15 - 22

Published: Jan. 1, 2025

Abstract Aim: This review examines the transformative potential of artificial intelligence (AI) in forensic science, emphasizing its applications crime scene analysis, evidence interpretation, digital forensics, and odontology. It highlights AI’s ability to enhance accuracy, efficiency, reliability while addressing ethical practical challenges. Materials Methods: A systematic search was conducted across PubMed, Web Science, Scopus, Google Scholar, complemented by manual reviews key journals grey literature. The included studies on AI odontology other domains published past decade. Predefined inclusion exclusion criteria were applied, duplicates removed. Full-text ensure relevance, with disagreements resolved through consensus a third reviewer rigor. Results: has significantly enhanced practices automating analysis improving accuracy. streamlines reconstruction, accelerates processes analyzing large datasets, advances dental forensics rapid victim identification bite mark analysis. AI-powered biometric systems suspect facial recognition pattern-matching technologies. However, limitations such as algorithmic bias, data privacy issues, resource disparities pose challenges widespread adoption. Conclusion: is revolutionizing science providing precision, investigations. Addressing concerns transparency, fairness, accountability crucial for responsible implementation. Future advancements should prioritize development explainable unbiased algorithms, privacy-preserving techniques, frameworks. Interdisciplinary collaborations global policy guidelines are essential equitable integration ultimately advancing justice equity criminal system.

Language: Английский

---

Personalized Financial Services through NLP and AI-Driven Innovations in FinTech

IGI Global eBooks, Journal Year: 2025, Volume and Issue: unknown, P. 371 - 420

Published: April 25, 2025

Individualized financial services are becoming more prevalent in the rapidly changing FinTech environment because to attract & retain clients alike. The effect of NLP AI-driven solutions is profound how institutions interact with their customers. This chapter examines AI personalized service delivery within industry. It looks at real-world applications that highlight customer communications through NLP-driven chatbots, virtual assistants, recommendation engines meant for providing tailored investment strategies, advice on an individual basis or assistance. Together, data privacy, algorithmic bias, regulatory compliance remain among toughest challenges faced by systems hence this provides insights into can utilize them morally without contravening any laws place. On one hand, it opportunities associated AI-based personalization while exploring its inherent risks thus giving a clear picture future engagement services.

Language: Английский

---

Mitigating Bias in Machine Learning Models with Ethics-Based Initiatives: The Case of Sepsis

The American Journal of Bioethics, Journal Year: 2025, Volume and Issue: unknown, P. 1 - 14

Published: May 12, 2025

This paper discusses ethics-based strategies for mitigating bias in machine learning models used to predict sepsis onset. The first part how various kinds of and their potential synergies can reduce predictive accuracy, especially as those biases derive from social determinants health (SDOHs) the design construction model. second essay certain ethically-based might mitigate disparate or unfair treatment produced by these models, not only they apply but any syndrome that witnesses impact adverse SDOHs on socioeconomically disadvantaged marginalized populations.

Language: Английский

---

Algorithmic Bias and Fairness in Biomedical and Health Research

Advances in computational intelligence and robotics book series, Journal Year: 2025, Volume and Issue: unknown, P. 287 - 324

Published: May 14, 2025

The rapid integration of artificial intelligence (AI) and machine learning (ML) into biomedical health research has the potential to transform patient care, diagnosis, treatment outcomes. However, as these technologies evolve, concerns surrounding algorithmic bias fairness have emerged. In context healthcare, biased algorithms can exacerbate disparities in outcomes, leading inequality care undermining trust AI-driven systems. This chapter explores ethical implications research, focusing on factors contributing datasets, model design, decision-making processes. Additionally, it examines various strategies frameworks aimed at promoting equity AI applications. Through a multidisciplinary lens, presents critical analysis how be achieved, with particular emphasis practical solutions regulatory considerations safeguard both integrity well-being diverse populations

Language: Английский

---

Large language models in biomedicine and health: current research landscape and future directions

Journal of the American Medical Informatics Association, Journal Year: 2024, Volume and Issue: 31(9), P. 1801 - 1811

Published: July 16, 2024

Large language models in biomedicine and health: current research landscape future directions (LLMs) are a specialized type of generative artificial intelligence (AI) focused on generating natural text.These developed through extensive training massive amounts text data use deep learning algorithms to generate new that closely resembles human-generated text.Generative AI methods, including LLMs, rapidly transforming various domains, healthcare. [1]2][3][4][5][6] They have already demonstrated remarkable potential as means process analyze large text, interpret language, content these domains.For example, Nori et al reported GPT-4 is able correctly answer the majority questions from medical practice licensing exams, comfortably obtaining passing grade. 7 Similarly, Stribling found this model exceeded average performance students graduate sciences examinations, strong short essay questions. 8 Even though exam not same applying knowledge real-world setting, results demonstrate LLMs can appropriate multiple-choice narrative responses framed language.ChatGPT, first released November 2022, has garnered phenomenal attention both scientific community broader society.A keyword search "large models" OR "ChatGPT" PubMed returned over 4500 articles discuss technology its implications for topics, informatics, by end June 2024.In addition, LLM-based technologies been deployed several healthcare systems offered integrated products clinic within vendor electronic health record (for thoughts initial evaluations an early product, see Garcia 9 Tai-Seale 10 ).This rapid adoption like ChatGPT brings unprecedented opportunity novel transform medicine.2][13][14][15][16][17] With great also comes need trustworthy responsible development technology.As we continue explore capabilities other it critical address related ethical, legal, social issues ensure used ways safe, fair, trustworthy, beneficial all.In context healthcare, particularly important engage stakeholders, such researchers, developers data-driven clinical decision support, care providers, system implementers academic centers industry, good.To accelerate area, issued call submissions Summer 2023, specifically focusing intersection biomedicine/health invited contributions all aspects.We report innovative informatics methods evaluation, well studies effectiveness/limitations methodologies healthcare.We encouraged challenges opportunities offer insights into how fields work together advance healthcare.This editorial provides overview papers accepted Focus Issue.We highlight major themes unique aspects ongoing challenges, recommend directions.Box 1 lists relevant terms abbreviations editorial. Overall statistics IssueThis JAMIA Issue drawn enthusiasm many researchers across different disciplines.In total, received 150 authors 25 countries regions 6 continents worldwide.The rigorous peer review was applied submissions, 41 which were ultimately publication (Table 1).The authored North America, followed those Asia Europe (Figure 1A).The highlights nature multi-disciplinary collaboration broad community.The number per paper varies 23, with 7.3.Many feature diverse expertise departments organizations.The authors' spans wide range fields, computer science, statistics, medicine, nursing, services, public policies, more.Several collaborations sectors, academia, government labs, institutes, hospitals, industry.Additionally, few showcase international among authors.

Language: Английский

---