Artificial Intelligence in Multidisciplinary, Schémas de Droit

Télécharge Artificial Intelligence in Multidisciplinary et plus Schémas au format PDF de Droit sur Docsity uniquement!

Critical Appraisal of Graumann et al. (2025): Artificial Intelligence in Multidisciplinary Ultrasound Applications Name: ID:

Critical Appraisal of Graumann et al. (2025): Artificial Intelligence in Multidisciplinary Ultrasound Applications

1. Introduction Ultrasound imaging has become a cornerstone of diagnostic medicine owing to its portability, safety, and real-time capabilities. Yet interpretation remains highly operator- dependent and susceptible to variability, particularly in complex anatomical regions such as the abdomen, musculoskeletal system, and obstetrics. With advances in computing power and deep learning, artificial intelligence (AI) has emerged as a transformative tool for enhancing diagnostic accuracy, automating image analysis, and improving workflow efficiency in sonography. The integration of AI into ultrasound practice supports precision medicine by offering reproducible and quantitative assessments (Getzmann et al., 2024). The rapid growth of AI-based image interpretation underscores the need for critical evaluation of methodological quality and clinical applicability. Graumann et al. (2025) provide one of the most comprehensive, multidisciplinary investigations of AI use across abdominal, gynecological, obstetric, musculoskeletal, vascular, and interventional ultrasound. Given the breadth and novelty of this publication, a structured appraisal is essential to determine the robustness, reliability, and ethical soundness of its conclusions. This critical appraisal applies the Critical Appraisal Skills Programme (CASP) checklist for systematic and narrative reviews (CASP, 2023) to evaluate the design, methodology, sample selection, data integrity, validity, reliability, and overall contribution of Graumann et al. (2025) to evidence-based radiology.

(AUC) values drawn from the reviewed studies. The article also highlighted technical limitations, such as small training datasets and lack of multicenter validation. The findings revealed that AI systems consistently achieved diagnostic accuracy comparable to or exceeding experienced radiologists for well-defined imaging tasks. AI also demonstrated potential for reducing examination time, enhancing image quality, and providing decision support during interventional procedures. Nonetheless, the authors cautioned that limited dataset diversity, insufficient regulatory guidance, and ethical challenges remain major obstacles to clinical translation.

3. Critical Analysis (CASP Framework) a. Study Design and Methodology The article adopts a narrative review design enriched by structured elements resembling a systematic approach. According to the CASP checklist, a well-conducted review should clearly define its purpose, justify the review type, and transparently describe inclusion criteria. Graumann et al. (2025) satisfy the first two criteria but only partially meet the third. While they specify databases (PubMed and Scopus) and temporal range (2015–2025), they do not provide a detailed search strategy for instance, the use of Boolean operators, inclusion/exclusion filters, or language restrictions. The authors present a multidisciplinary perspective, drawing on clinical, engineering, and ethical expertise. This breadth compensates for the lack of strict systematic rigor, ensuring that findings reflect real-world diversity across ultrasound specialties. However, unlike Getzmann et al. (2024) who employed a fully systematic approach with PRISMA diagrams and bias assessment Graumann et al. (2025) focus on conceptual integration rather than quantitative validation.

Their methodology remains appropriate for an emerging technology field where heterogeneity among primary studies precludes formal meta-analysis. Yet, explicit reporting of data extraction methods and study-quality appraisal tools (e.g., JBI or ROBIS) would have enhanced reproducibility and credibility. b. Sample Selection and Size The review encompasses more than 160 original studies, offering a broad evidence base that supports its generalizability. These studies span diverse medical contexts—ranging from fetal imaging to vascular assessment representing the largest multidisciplinary synthesis in AI ultrasound to date. However, the selection process lacks transparency. The absence of a flow diagram or numerical breakdown of excluded articles prevents assessment of potential selection bias. For instance, studies demonstrating weak AI performance or negative outcomes might have been underrepresented, inflating perceived algorithmic efficacy. CASP highlights that lack of clear sampling criteria can compromise interpretive reliability. Nevertheless, the review’s inclusiveness is a significant strength. By integrating evidence across multiple organ systems and imaging objectives, the authors capture a holistic picture of AI’s capabilities. This diversity enhances external validity, reflecting the multifaceted nature of modern ultrasound practice. As noted by Dubey et al. (2025), comprehensive sampling is essential to evaluate AI’s impact on procedural guidance and interventional safety. c. Data Collection and Analysis Methods Data collection procedures were systematically organized by organ category, summarizing algorithm types (CNNs, U-Net architectures, random-forest classifiers) and

Reliability. Findings align with parallel evidence: Getzmann et al. (2024) confirmed that machine-learning algorithms can match expert accuracy for tendon and nerve evaluation, supporting Graumann et al.’s broader conclusions. The convergence of outcomes across independent reviews enhances reliability. e. Strengths and Limitations Strengths  Comprehensiveness: Inclusion of six subspecialties positions this as one of the most wide-ranging AI–ultrasound reviews to date.  Interdisciplinary perspective: Collaboration among radiologists, obstetricians, vascular specialists, and engineers enhances interpretive depth.  Clinical relevance: Real-world scenarios, such as AI guidance in interventional procedures and obstetric measurements, bridge laboratory research with clinical workflows.  Forward-looking analysis: Discussion extends to regulatory approval, ethics, and integration with Picture Archiving and Communication Systems (PACS). Limitations  Lack of systematic search transparency: No PRISMA diagram or explicit inclusion/exclusion checklist.  Absence of quantitative synthesis: Meta-analytic validation of pooled performance metrics is missing.

 Potential publication bias: Positive findings likely overrepresented, as negative AI trials are rarely published.  Limited discussion on data sovereignty: Ethical implications of patient data transfer for AI training require deeper exploration. These limitations are consistent with early-phase technology reviews, where heterogeneity across algorithms and imaging protocols complicates quantitative aggregation. f. Bias and Ethical Considerations The authors disclosed no conflicts of interest, supporting transparency. Ethical considerations are addressed conceptually, noting challenges in data privacy, informed consent for AI training datasets, and algorithmic fairness. However, CASP encourages explicit discussion of governance mechanisms and bias mitigation strategies, which remain underdeveloped in the paper. Algorithmic bias especially gender, ethnicity, or equipment bias was acknowledged but not quantitatively analyzed. This oversight parallels findings by Dubey et al. (2025), who highlighted that limited dataset diversity may perpetuate inequities in interventional outcomes. Furthermore, Graumann et al. emphasized the need for explainable AI models to enhance clinician trust, an ethical imperative in diagnostic decision-making. g. Applicability and Clinical Impact From a clinical translation perspective, the article provides actionable insights. It demonstrates how AI can:

1. Automate fetal biometric measurements to reduce inter-observer variability;
2. Assist in musculoskeletal lesion classification and segmentation;
3. Enhance vascular flow quantification; and
4. Guide needle positioning in interventional ultrasound.

5. References CASP. (2023). Critical Appraisal Skills Programme (CASP) Checklist for Systematic Reviews. Available at: https://casp-uk.net/casp-tools-checklists/ [Accessed 10 Nov 2025]. Dubey, A., Uldin, H., Khan, Z., Panchal, H., Iyengar, K., & Botchu, R. (2025). Role of Artificial Intelligence in Musculoskeletal Interventions. Cancers , 17, 1615. https://doi.org/10.3390/cancers Getzmann, J., Zantonelli, G., Messina, C., Albano, D., Serpi, F., Gitto, S., & Sconfienza, L. (2024). The use of artificial intelligence in musculoskeletal ultrasound: A systematic review of the literature. La Radiologia Medica , 129, 1405–1411. https://doi.org/10.1007/s11547-024-01856- Graumann, O., Xin, W., Goudie, A., Blaivas, M., Braden, B., Westerway, S., Chammas, M., Dong, Y., Gilja, O., Hsieh, P., Tian, A., Liang, P., Möller, K., Nolsøe, C., Saftoiu, A., & Dietrich, C. (2025). Artificial Intelligence in Abdominal, Gynecological, Obstetric, Musculoskeletal, Vascular and Interventional Ultrasound. Ultrasound in Medicine & Biology. https://doi.org/10.1016/j.ultrasmedbio.2025.07.