Artificial Intelligence in Radiology: Transforming Diagnostics and Raising Ethical Dilemmas

Abstract

Background: Artificial intelligence (AI) is revolutionizing radiology by improving image analysis, enhancing diagnostic accuracy, and streamlining workflows. Deep learning algorithms, especially convolutional neural networks (CNNs), have shown better performance in lesion detection, classification, and quantification. Challenges to implementing AI in clinical practice include ethical issues, regulatory barriers, and the requirement for strong validation.Aim: The current uses, advantages, and shortcomings of AI in radiology are reviewed, specifically image interpretation, workflow enhancement, and clinical decision support. In addition to the above, this review considers ethical issues, regulations, and prospective directions in adopting AI in radiology.Material and Method: Literature search using PubMed, PMC, and other biomedical databases was performed. Research on AI applications in radiology, encompassing diagnostic accuracy, workflow efficiency, and ethical/legal hurdles, was reviewed. Case studies, clinical trials, and meta-analyses were included to determine real-world performance and barriers to adoption.Results: AI has demonstrated considerable potential to enhance diagnostic accuracy (e.g., false positives in mammography by as much as 83%) and workflow efficiency (e.g., reporting times in emergency chest X-rays by as much as 77%). Variation between AI models exists, as well as an impact of bias in training data on performance. Ethical issues, including algorithmic bias and patient privacy, are still unanswered. Regulatory environments (FDA, CE marking) are changing, but legal responsibility for AI-aided diagnoses remains primarily with radiologists.Conclusion: AI has tremendous potential to improve radiology by making it more efficient and diagnostic accurate. Successful implementation, however, needs to overcome ethical, legal, and technical hurdles. Future developments must focus on explainable AI, standardized validation, and multidisciplinary collaboration to provide equitable and effective deployment.

Keywords

Radiology, artificial intelligence, deep learning, workflow optimization, ethical considerations, regulatory challenges, diagnostic accuracy

Introduction

Applications of AI in Radiology

AI in Image Interpretation

AI performs very well on fundamental radiology tasks of detection, classification, and quantification. Contemporary CNNs are able to distinguish normal vs. abnormal structures with high accuracy(7). For instance, deep-learning models have been designed to separate benign from malignant results in mammography with performance on par with human radiologists(14). These image-classification programs are computerized diagnostic tools (frequently referred to as computer-aided detection/diagnosis, or CAD) that mark out suspicious areas for radiologist review(15). CNNs are also employed to segment lesions (e.g. tumors) and organs, allowing for automated volume or shape change measurements over time(16). Such measurement can facilitate therapy monitoring: for example, AI-segmented tumor volumes from MRI have been found to correlate with prostate patient outcomes cancer(17). AI-powered CAD systems have reported remarkable decreases in false positives. In one case, an AI-CAD system beat an older CAD software by decreasing false-positive marks per image by 69%(18). It detected microcalcifications and masses in mammograms with a much smaller number of false alarms (reducing false positives by 83% and 56%, respectively), which can reduce reading time and lower radiologist fatigue(19). Consequently, AI as a "second reader" can enhance diagnostic quality while permitting radiologists to concentrate on actual abnormalities(20). Therefore, AI as a "second reader" can improve diagnostic quality while allowing radiologists to focus on real (achieving area-under-curve ≈0.93)(21). Even with these advancements, research also points to differences among AI instruments. Kondrashova et al. contrasted two FDA-approved AI algorithms for detecting and quantifying lung nodules in 946 screening CT scans(22). They discovered highly significant correlation between measured nodule volumes (r>0.95) but with a minor systematic difference: one device systematically reported higher volumes than the other (mean difference ~6?mm³)(23). This volume discord drove discordant Lung-RADS classification in 38% of patients, which could change patient management(24). These examples highlight that AI responses are not interchangeable; the training data and methodology of each model influence its performance and thresholds(25). Careful clinical validation is thus necessary prior to deployment of any AI interpretation tool.

Workflow Optimization

AI can optimize radiology workflows through automation of standard steps, prioritization of cases, and minimized manual effort. Key examples include triaging of imaging studies and report generation via automation. For example, AI systems can pre-screen incoming studies to alert on emergencies. In one real-world study, an FDA-cleared AI solution (Lunit CXR Triage) was implemented in an emergency chest X-ray workflow(8). Within a span of five months (20,944 cases), the AI realized 99% specificity in highlighting urgent findings, and through putting spotlight on most critical studies it is estimated to have minimized reporting turnaround times by 77%(27). Practically, such triage instruments enable radiologists to deal with life-or-death situations (e.g. pneumothorax, occlusion of the great vessels) more rapidly, enhancing patient outcome.

Triage and Prioritization: AI may categorize scans (e.g., as normal, not urgent, or urgent) prior to review by radiologists. On the above chest X-ray exam, 10.2% of them were marked "urgent," and the high level of specificity (99%) ensured radiologists could rely on the AI for prioritizing urgent reports(28). Likewise, artificial intelligence algorithms for stroke imaging now automatically identify acute infarct or hemorrhage signs to expedite treatment decisions(29). These systems also function as "first readers" which re-prioritize worklists or send out reminders for high-risk findings(12).

Secondary Reading and Verification: In other processes, AI serves as a "second reader." One systematic review indicated that in most applications, AI is used subsequent to a human read to ensure that there was no missed findings(12). This method is able to catch misses (e.g. minor nodules) and offer a safety net, although the added step has to be weighed against possible increases in review time.

Time and Workload Reduction: By and large, the majority of research on AI in use documents gains in efficiency. Across meta-analysis, 67% of task time-measuring studies concluded that AI use resulted in decreased reading time or processing time(30). Nevertheless, pooled analyses had heterogeneity: certain meta-analyses did not reveal a statistically significant total effect on time saving due to variations in study design and integration approaches. In spite of conflicting data, the trend is auspicious: AI frees radiologists from routine tasks (such as lesion measurements), and AI-aided reading has been linked with approximately 17% decrease in case-reading time in controlled studies(31).

Reporting Automation: AI also has potential for semi-automated reporting. Natural language processing (NLP) can be used to extract structured findings from previous reports or fill in templates according to image characteristics, reducing dictation. Fully AI-generated reports are not yet routine, but commercial products are in development that will write preliminary report text or provide suitable follow-ups. Incorporation of AI findings into reporting (e.g. auto-annotated images) may further accelerate documentation and normalize reporting language(32).

Clinical Decision Support Systems

In addition to image interpretation, AI can be used as a decision support system that combines imaging with clinical information to inform diagnosis and management. Several AI applications already offer clinical guidance:

Computer-Aided Diagnosis (CAD): Older CAD technologies, like computer-aided detection of lung nodules or mammographic calcifications, enhance sensitivity by pointing to subtle findings. More recent deep-learning CAD software builds on this by both detecting and characterizing lesions (e.g. providing probability of malignancy suggestions). Such suggestions can help radiologists make differential diagnoses(9).

Prognostic and Predictive Analytics: AI can quantify imaging biomarkers (radiomics) that are correlated with outcomes. For instance, AI-segmentation of tumor volumes has independently predicted progression-free survival in cancer patients. By examining patterns between imaging, pathology, and genomics, AI-driven radiomic models hope to individualize risk assessment and treatment planning(33). Such algorithms may, for example, inform us which nodules in the lungs require biopsy vs. follow-up, or which breast cancers will be sensitive to treatment.

Standardized Scoring: AI supports scoring systems that inform management. In the setting of stroke, for example, computerized software may calculate the ASPECTS (Alberta Stroke Program Early CT Score) to measure infarct size, assisting emergency doctors in making intervention decisions. Likewise, breast imaging AI may automatically classify BIRADS categories and suggest follow-up. Automated scores create consistency and assist with ensuring compliance with evidence-based guidelines(34).

Integrating Multimodal Data: A few next-generation AI systems integrate imaging with electronic health records (labs, genomics) to further refine decision-making. For instance, a radiology-AI system could integrate CT results with cancer biomarkers to suggest additional testing or treatment. While numerous such systems are still in development stages, initial research indicates they have the potential to enhance diagnostic accuracy and support development of patient-specific care plans. In all these jobs, AI assists the clinician's decision but is designed to operate in parallel with – and not in lieu of – human judgment. According to one review, AI may "augment clinical decision-making, enabling radiologists to dedicate more time to more challenging diagnosis challenges(35).

Ethical Considerations

The application of AI in radiology poses significant ethical concerns. The major issues are algorithmic fairness and bias, transparency/explainability, and data privacy/patient consent.

Algorithmic Bias and Fairness: AI algorithms can reflect and even magnify biases that are in the data used for their training. As an example, if the dataset underrepresented a patient subgroup (e.g., a certain race or age), the AI would perform worse on those patients. Bias also emerges from the practices of annotating: a potential example is a mammography AI trained from radiologist annotations which prioritized malignant masses over benign calcifications(10). If benign results were not always marked, the AI could become "biased" towards identifying cancers and ignoring benign illnesses. Likewise, "automation bias" can happen if radiologists over-rely on AI results; for instance, a radiologist might miss pneumonia signs if an AI algorithm was trained to look for nodules, thus missing the correct diagnosis. These biases compromise equity: they can cause misdiagnoses, unequal access to diagnosis, and poorer outcomes for underrepresented patients. Multiple approaches must be used to mitigate bias. The literature stresses employing large, diverse, and representative datasets in model development(36). For example, China's AI guidelines specifically promote training on data from several hospitals and demographic groups to prevent overfitting to a single population. Independent audits and post-deployment monitoring can also identify performance differences. Multidisciplinary team involvement (data scientists, clinicians, ethicists) is suggested to review AI behavior. Finally, fairness is a fundamental AI ethic: "AI biases and discrimination" in medical AI are identified as serious issues that might cause harm to patient health. Policy frameworks such as the envisioned European "AI Act" also emphasize the necessity of fairness and non-discrimination in AI systems(37).

Transparency and Explainability: Most AI algorithms, particularly deep neural networks, are "black boxes" whose internal rationality is difficult to interpret. Such transparency can lead to the erosion of trust by clinicians and patients. A radiologist, for instance, may hesitate to take action on a cancer diagnosis by an AI if he or she cannot understand why the AI concluded so. To this end, regulators and researchers promote "explainable AI" (XAI) approaches that uncover why specific image features propel the output. Ethical standards call for AI systems ideally to explain their reasoning in terms comprehensible by humans. Explainability also intersects legal accountability: if an AI suggestion results in harm to patients, clinicians and jurists will be curious about the justification. Explainability is hard but important in medicine. There are some solutions like heatmaps displaying which pixels contributed to the decision of AI, or the use of easier models (when feasible) to permit logical reasoning. As per one review, "ensuring transparency in how AI algorithms work" and "developing explainable AI systems" are of vital importance in ensuring safe inclusion into radiology. Transparency also extends to algorithm development: ideally, vendors would share performance data, limitations, and version updates with users such that radiologists are aware of the strengths and limitations of the tool they use.

Data Privacy and Patient Consent: AI depends on huge sets of patient images and data, with potential privacy implications. Patient health data (PHI) need to be secured by regulations such as HIPAA (US) and GDPR (EU). But certain AI methods (e.g. generative models) theoretically have the capacity to re-identify anonymized images and disclose sensitive information. Such "de-anonymization" risk implies that even de-identified imaging data sets need stringent protections. Ethicists caution that regulatory control tends to trail behind technology; an analysis underscores that "regulation should emphasize patient agency and consent" and promote new anonymization techniques(3). In practice, this entails that patients should in general agree to their images being used for AI training or testing, and hospitals should impose robust encryption and access controls. In addition, international variations in data-protection regulations complicate things. The EU's GDPR deems health data to be particularly sensitive and requires formal consent and restricts cross-border transfers. HIPAA regulates the use of medical data in the United States but doesn't explicitly mention AI. Radiology practices that implement AI have to work around these systems. In certain applications (e.g. public health studies of AI), waivers by ethics boards or broad consent may be sufficient, but patient transparency is essential. Ultimately, data governance challenges intersect with fairness: if databases omit groups for privacy reasons, that in and of itself may create bias.

Legal and Regulatory Challenges

Legislative and regulatory environments for AI in radiology continue to emerge. Of principal concern are responsibilities for mistakes, mechanisms for approving devices, and differences in international law.

Accountability and Liability: Who is liable if an AI-aided diagnosis is incorrect? Practice currently tends to put responsibility on the human clinician. Radiologists are required to check and confirm AI results, and can still be held liable for negligence if an AI error causes harm. In malpractice law, the radiologist's responsibility still applies: if a radiologist unquestioningly adopts an AI report without doing the standard of care (e.g. seeing an obvious tumor the AI did not catch), the radiologist may be held negligent(38). Surveys of radiologists confirm this expectation: in one survey in Europe, 45% of those surveyed reported that radiologists should be responsible for any output of AI that affects clinical decisions. While this is happening, hospital systems and developers might be liable under alternative theories of law. Hospitals might have "vicarious liability" if an AI use by an employee causes damage. Software developers may be subject to product-liability suits if their AI is considered a faulty device, although so far there are not many instances that have directly tackled this. Legal scholars suggest that as autonomy increases in AI, available structures (medical malpractice, product liability, negligence) might require updating. Others have proposed unorthodox remedies such as AI "personhood" or exceptional compensation regimes, but these remain speculative. Practically, because there is currently no clear AI-specific case law, legal accountability will most likely be determined case by case. Experts advise that developers clearly document the intended function and bounds of AI tools (similar to medicine insert leaflets) to specify liability(38).

Regulatory Frameworks and Approval Pathways: In most nations, AI radiology software is governed as a medical device. In the United States, the Food and Drug Administration (FDA) examines AI/ML-powered medical devices, usually subject to 510(k) clearance, de novo approval, or Pre-Market Approval (PMA) based on risk level(39). Traditionally, the majority of FDA-approved AI imaging devices have come through the 510(k) route (proving substantial similarity to a predicate device)(36). For instance, out of 521 FDA-cleared AI medical devices that have been identified up to mid-2023, 96% were cleared under 510(k), while a mere 18 were de novo and only 3 received PMA. Importantly, all these were marked as Class II devices in the US, representing a moderate risk categorization. The FDA also calls for continued monitoring of AI tools, and has put forth a "total product lifecycle" model for algorithms that learn progressively, to secure safety through updates. For example, of 521 FDA-approved AI medical devices identified through mid-2023, 96% were approved under 510(k), and just 18 were de novo, with only 3 receiving PMA. Crucially, all of these were labeled as Class II devices in the US, which is a moderate risk designation. The FDA also demands ongoing surveillance of AI tools, and has proposed a "total product lifecycle" model for progressively learning algorithms, to achieve safety through updates seen(1). Other jurisdictions have different procedures. China's National Medical Products Administration (NMPA) has approved dozens of AI imaging products to date (2023), but whereas the US/EU considers new AI software (without previous approval) as low-risk, China classifies new AI software (no prior approval) as high-risk (Class III) devices. China's policy is "rule-based": very detailed guidelines define how manufacturers are required to report the whole lifecycle of AI development (data, algorithm, testing) and in effect produce a "digital twin" of the algorithm for regulators. US FDA takes more outcome-based standards and provides manufacturers with flexibility in implementation of safety. Canada is slowly evolving its regulatory position, and most countries adopt mixes of these models(36).

International Variations: Since laws vary, an AI radiology device might have CE marking in Europe but not FDA clearance in the US (or vice versa). For instance, Lunit's CXR Triage is FDA-cleared within the US; it also has CE marking for Europe. Some firms first obtain FDA clearance (considered a "gold standard") and later sell the same technology in other markets. But there are discrepancies: China insists on local clinical testing for each product, whereas the US/EU recognize foreign study data on certain conditions. New global policy initiatives, including the EU AI Act or WHO guidance on AI in healthcare, seek to standardize some points (e.g. transparency, risk management), but national implementation will persist. Radiology clinics that use AI will thus have to operate across several legal regimes, being in compliance with each relevant regulatory standard and tracing the provenance of their AI software.

Current Research and Case Studies

The area of AI in radiology is moving very fast, with new deployments, tools and studies appearing. Some examples below show recent outcomes and trends:

Real-World Deployment Studies: In practice, validation studies have started measuring the effect of AI. The Lunit chest X-ray triage trial (2024) is a flagship. Integrated into a PACS of a hospital, it worked through more than 20,000 emergency CXRs and showed high performance: 99% specificity for urgent results and an 89% sensitivity to detect normal scans(28). The authors observed that by giving precedence to priority cases, the tool could significantly limit reporting delays. In another meta-analysis by 48 AI clinical imaging efficiency studies were surveyed. It observed workflow adjustments entailed AI as a secondary reader (reviewing all images) or a primary screener (isolating positives). A total of 67% of studies reported time gains in task completion, although pooled outcomes were heterogeneous. Such studies highlight the capabilities and limitations of AI in everyday workflows(12).

Diagnostic Imaging Trials: Clinical testing is also evaluating diagnostic performance. For example, compared two computer-aided detection (CAD) algorithms for lung cancer screening (the HANSE trial). Although the two algorithms both had high correlation in nodule volume measurements (r>0.95), large differences in measurements resulted in discordant management recommendations (38% of cases had different Lung-RADS categories). This indicates a need for standardization between products. Other studies have focused on AI use in mammographic screening (which is not covered in this), intracranial hemorrhage detection, and tuberculosis detection from X-rays, usually demonstrating the ability of high-performing AI models to approximate radiologist outcomes when tested over large datasets(40).

Emerging Research Topics: New studies persist in radiomics and multimodal AI. Research is assessing the predictive value of AI-extracted imaging features for treatment response or genetic biomarkers. For instance, radiomic assessment of multiparametric MRI in prostate cancer indicated that AI-segmented tumor volumes were independent prognostic factors for recurrence(17). Work is also being done in combining image data with genomic and clinical data to construct comprehensive models for precision. oncology(31). Outside of oncology, work on AI for neurological disease (e.g., Alzheimer's diagnosis on MRI) and cardiology (e.g., calcium scoring) is underway.

Stakeholder Surveys and Attitudes: Surveys indicate how radiologists perceive AI in practice. A 2024 European survey (ESR/EuroAIM) reported that almost half of radiologists already had AI in their workflow, with specific impact anticipated in breast and oncologic imaging. Significantly, the survey showed that 45% of radiologists think they will still be held accountable for decisions made with AI, showing wariness regarding liability. A further survey revealed that the majority of radiologists would not approve completely AI-produced reports with no human oversight. Such a finding reveals that clinicians acknowledge the advantages of AI (e.g. efficiency, assistance in detection) but are also aware of its limitations and the necessity of human governance(11).

DISCUSSION

Opportunities

AI presents numerous potential advantages for radiology. By enhancing detection sensitivity and decreasing workload, AI can potentially alleviate the long-term shortage of radiologists globally. Force multiplier tools can be provided by automated systems – one radiologist can manage significantly more studies with AI support. In screening programs (e.g. lung cancer screening CT, breast mammography), AI can sort out negatives so that reading numbers are decreased, radiologist effort concentrated on probable positives. In low-resource areas, cloud-based AI may offer initial reads where professionals are scarce(13). Apart from efficiency, AI may improve patient outcomes. Early studies suggest that AI can identify fine details humans may overlook (e.g. minute lung nodules or incipient bone fractures) and measure disease progression more quantitatively. In acute conditions such as stroke or trauma, prompt AI notification may accelerate life-saving interventions. AI also facilitates radiomics – integration of imaging with multi-omic analyses – which could usher in genuinely personalized diagnostics and prognostics. If designed carefully, AI can release radiologists from routine tasks (e.g. measurements, report generation), enabling them to concentrate on difficult cases, interdisciplinary management and patient communication. In such a vision, radiologists become expert overseers or "validators" of AI results. Partnership with AI engineers and data scientists can also fuel innovation, creating tools optimized to real clinical needs(41).

Limitations and Challenges

In spite of promise, radiology AI has several major limitations. Biases and data flaws (as above) can curtail real-world performance, and most published AI algorithms do not generalize across diverse patient populations or imaging hardware. Regulatory barriers translate to the need for rigorous clinical validation prior to adoption, and this can slow deployment. Technical complexity in integration into existing hospital IT systems (PACS, RIS, EHR) can require interoperability standards still in development. Human factors are also not trivial. Radiologists can become fatigued by alerts if AI generates too many false alarms, or disbelieve AI if it is too often incorrect. In fact, some reports suggest AI can even decrease a radiologist's accuracy slightly if it gives erroneous cues. To make effective use of AI will likely involve training and workflow adjustment. Moreover, medico-legal issues will lead institutions to be cautious against depending too much on AI. The expense of AI solutions is also a factor. AI companies tend to have licensing or per-study pricing, and the hospitals need to balance these against unproven efficiency gains. The business case for AI still needs to be clearly established in most settings.

Future Directions

Research and policy are catching up in a hurry. Methodologically, the future wave of AI research prioritizes "explainable AI", strong external validation, and adaptive learning systems that refine their performance as more data become available. The FDA and other regulators are debating "real-world performance monitoring" as a condition of approval, to guarantee AI continues to be accurate after deployment. Efforts are being made across disciplines to harmonize reporting of AI validation studies (e.g. CLAIM checklist) and formulate international guidelines for algorithm explainability and bias auditing. At the clinical level, some possible areas of development are: not just integration of AI in radiology departments but throughout medicine (AI alerts within general practice or emergency workflows), unified collaboration of AI with other diagnoses (bringing together imaging AI with genomic AI), and the leveraging of large language models (LLMs) for radiologist use in report production and retrieval of information. New technologies such as federated learning can assist hospitals in training AI on multi-site data without ever sharing raw patient images, allaying privacy concerns. Policy makers are especially key stakeholders. To capture the benefits of AI, they must create innovation-friendly environments without endangering patients. This could include more transparent liability rules (e.g. shared responsibility models), data sharing funds as incentives, and public-private collaborations for building equitable datasets. Radiologists need to play an active role – defining clinical requirements for AI, working in development and testing, and assuring that they have consideration taken into their policies. Radiologist, developer, and regulator partnerships – as others point out – will be key in order to ensure a safe and effective adoption of AI.

CONCLUSION

Artificial intelligence has the potential to revolutionize radiology by improving diagnostic accuracy and efficiency, but this revolution is accompanied by ethical and regulatory challenges. Recent research illustrates that AI can assist image interpretation (matching or outperforming human performance in most instances) and can optimize workflows (e.g. triaging high-risk cases). Concurrently, AI presents significant ethical issues: algorithmic bias risks undermining fairness, black-box models risk undermining trust, and the use of patient data for AI necessitates regard for privacy and consent. Legally, accountability is still an open issue, with radiologists typically responsible for monitoring AI outputs, while regulatory agencies are adjusting to determine how to categorize AI instruments as medical devices (through FDA approvals, CE marks, etc.). Globally, methods range from the US/EU standards-based systems to China's rules-based precise process. In the future, the challenge will be innovating while being cautious. Developers of AI must put fairness, transparency and evidence at the top of their products; radiologists must retain a critical watchful eye; and policymakers must develop regulations that guarantee safety without hampering innovation. All parties, including patients, must be involved in a dialogue. In the words of recent opinion pieces, AI in radiology is a powerful agent of change, but only if it is guided by "sustained innovation, dynamic partnerships, and a steadfast commitment to ethical responsibility". With such equilibrium paths, AI can enhance patient care, not by substituting for radiologists, but by making them able to provide more accurate diagnoses more quickly and more equitably than ever before.

REFERENCE

1. Obuchowicz R, Lasek J, Wodzi?ski M, Piórkowski A, Strzelecki M, Nurzynska K. Artificial Intelligence-Empowered Radiology—Current Status and Critical Review. Diagnostics. 2025 Jan 24;15(3):282.
2. Bhandari A. Revolutionizing Radiology With Artificial Intelligence. Cureus. 2024 Oct;16(10): e72646.
3. Murdoch B. Privacy and artificial intelligence: challenges for protecting health information in a new era. BMC Medical Ethics. 2021 Dec 15;22(1):122.
4. Nam JG, Hwang EJ, Kim J, Park N, Lee EH, Kim HJ, et al. AI Improves Nodule Detection on Chest Radiographs in a Health Screening Population: A Randomized Controlled Trial. Radiology. 2023 Apr 1;307(2).
5. Sah A kumar, Agarwal S, Abbas AM, Shalabi MG, Prabhakar PK, Elshaikh RH, et al. Advances in Image Processing and Pattern Recognition in Cancer Detection, Prediction, Diagnosis, and Prognosis. 2025.
6. Esmaeilzadeh P. Challenges and strategies for wide-scale artificial intelligence (AI) deployment in healthcare practices: A perspective for healthcare organizations. Artificial Intelligence in Medicine. 2024 May; 151:102861.
7. Hosny A, Parmar C, Quackenbush J, Schwartz LH, Aerts HJWL. Artificial intelligence in radiology. Nature Reviews Cancer. 2018 Aug 17;18(8):500–10.
8. Tyler S, Olis M, Aust N, Patel L, Simon L, Triantafyllidis C, et al. Use of Artificial Intelligence in Triage in Hospital Emergency Departments: A Scoping Review. Cureus. 2024 May;16(5): e59906.
9. Sharafaddini AM, Esfahani KK, Mansouri N. Deep learning approaches to detect breast cancer: a comprehensive review. Multimedia Tools and Applications. 2024 Aug 20;
10. Koçak B, Ponsiglione A, Stanzione A, Bluethgen C, Santinha J, Ugga L, et al. Bias in artificial intelligence for medical imaging: fundamentals, detection, avoidance, mitigation, challenges, ethics, and prospects. Diagnostic and interventional radiology (Ankara, Turkey). 2025 Mar 3;31(2):75–88.
11. Zanardo M, Visser JJ, Colarieti A, Cuocolo R, Klontzas ME, Pinto dos Santos D, et al. Impact of AI on radiology: a EuroAIM/EuSoMII 2024 survey among members of the European Society of Radiology. Insights into Imaging. 2024 Oct 7;15(1):240.
12. Wenderott K, Krups J, Zaruchas F, Weigl M. Effects of artificial intelligence implementation on efficiency in medical imaging—a systematic literature review and meta-analysis. npj Digital Medicine. 2024 Sep 30;7(1):265.
13. Katal S, York B, Gholamrezanezhad A. AI in radiology: From promise to practice − A guide to effective integration. European Journal of Radiology. 2024 Dec; 181:111798.
14. Li H, Zhuang S, Li D ao, Zhao J, Ma Y. Benign and malignant classification of mammogram images based on deep learning. Biomedical Signal Processing and Control. 2019 May; 51:347–54.
15. Boafo YG. An overview of computer—aided medical image classification. Multimedia Tools and Applications. 2024 Jun 17;
16. Lenchik L, Heacock L, Weaver AA, Boutin RD, Cook TS, Itri J, et al. Automated Segmentation of Tissues Using CT and MRI: A Systematic Review. Academic Radiology. 2019 Dec;26(12):1695–706.
17. Yang DD, Lee LK, Tsui JMG, Leeman JE, McClure HM, Sudhyadhom A, et al. AI-derived Tumor Volume from Multiparametric MRI and Outcomes in Localized Prostate Cancer. Radiology. 2024 Oct 1;313(1).
18. Mayo RC, Kent D, Sen LC, Kapoor M, Leung JWT, Watanabe AT. Reduction of False-Positive Markings on Mammograms: a Retrospective Comparison Study Using an Artificial Intelligence-Based CAD. Journal of Digital Imaging. 2019 Aug 8;32(4):618–24.
19. Clerkin N, Ski CF, Brennan PC, Strudwick R. Identification of factors associated with diagnostic performance variation in reporting of mammograms: A review. Radiography. 2023 Mar;29(2):340–6.
20. Robert D, Sathyamurthy S, Singh AK, Matta SA, Tadepalli M, Tanamala S, et al. Effect of Artificial Intelligence as a Second Reader on the Lung Nodule Detection and Localization Accuracy of Radiologists and Non-radiology Physicians in Chest Radiographs: A Multicenter Reader Study. Academic Radiology. 2025 Mar;32(3):1706–17.
21. RASOOL N, IQBAL BHAT J. Unveiling the Complexity of Medical Imaging through Deep Learning Approaches. Chaos Theory and Applications. 2023 Dec 31;5(4):267–80.
22. Liu JA, Yang IY, Tsai EB. Artificial Intelligence (AI) for Lung Nodules, From the AJR Special Series on AI Applications. American Journal of Roentgenology. 2022 Nov;219(5):703–12.
23. Guedes Pinto E, Penha D, Ravara S, Monaghan C, Hochhegger B, Marchiori E, et al. Factors influencing the outcome of volumetry tools for pulmonary nodule analysis: a systematic review and attempted meta-analysis. Insights into Imaging. 2023 Sep 23;14(1):152.
24. van Riel SJ, Jacobs C, Scholten ETh, Wittenberg R, Winkler Wille MM, de Hoop B, et al. Observer variability for Lung-RADS categorisation of lung cancer screening CTs: impact on patient management. European Radiology. 2019 Feb;29(2):924–31.
25. Liang W, Tadesse GA, Ho D, Fei-Fei L, Zaharia M, Zhang C, et al. Advances, challenges and opportunities in creating data for trustworthy AI. Nature Machine Intelligence. 2022 Aug 17;4(8):669–77.
26. Prateek M, Rathore SPS. Clinical Validation of AI Disease Detection Models — An Overview of the Clinical Validation Process for AI Disease Detection Models, and How They Can Be Validated for Accuracy and Effectiveness. In: AI in Disease Detection. Wiley; 2025. p. 215–37.
27. Shah H, Shah S, Tanwar S, Gupta R, Kumar N. Fusion of AI techniques to tackle COVID-19 pandemic: models, incidence rates, and future trends. Multimedia Systems. 2022 Aug 13;28(4):1189–222.
28. Will Morton. https://www.auntminnie.com/clinical-news/digital-x-ray/article/15705715/ai-performs-well-triaging-lung-xrays-in-realworld-settings#:~:text=AI,the%20European%20Journal%20of%20Radiology.
29. Soun JE, Chow DS, Nagamine M, Takhtawala RS, Filippi CG, Yu W, et al. Artificial Intelligence and Acute Stroke Imaging. AJNR American journal of neuroradiology. 2021 Jan;42(1):2–11.
30. Vaccaro M, Almaatouq A, Malone T. When combinations of humans and AI are useful: A systematic review and meta-analysis. Nature Human Behaviour. 2024 Oct 28;8(12):2293–303.
31. Najjar R. Redefining Radiology: A Review of Artificial Intelligence Integration in Medical Imaging. Diagnostics (Basel, Switzerland). 2023 Aug 25;13(17).
32. Wu C, Li X, Guo Y, Wang J, Ren Z, Wang M, et al. Natural language processing for smart construction: Current status and future directions. Automation in Construction. 2022 Feb; 134:104059.
33. Pang Y, Wang H, Li H. Medical Imaging Biomarker Discovery and Integration Towards AI-Based Personalized Radiotherapy. Frontiers in Oncology. 2022 Jan 17;11.
34. El Naamani K, Musmar B, Gupta N, Ikhdour O, Abdelrazeq H, Ghanem M, et al. The Artificial Intelligence Revolution in Stroke Care: A Decade of Scientific Evidence in Review. World Neurosurgery. 2024 Apr; 184:15–22.
35. Hirosawa T, Suzuki T, Shiraishi T, Hayashi A, Fujii Y, Harada T, et al. Adapting Artificial Intelligence Concepts to Enhance Clinical Decision-Making: A Hybrid Intelligence Framework. International Journal of General Medicine. 2024 Nov;Volume 17:5417–22.
36. Liu Y, Yu W, Dillon T. Regulatory responses and approval status of artificial intelligence medical devices with a focus on China. npj Digital Medicine. 2024 Sep 18;7(1):255.
37. Ueda D, Kakinuma T, Fujita S, Kamagata K, Fushimi Y, Ito R, et al. Fairness of artificial intelligence in healthcare: review and recommendations. Japanese journal of radiology. 2024 Jan;42(1):3–15.
38. Cestonaro C, Delicati A, Marcante B, Caenazzo L, Tozzo P. Defining medical liability when artificial intelligence is applied on diagnostic algorithms: a systematic review. Frontiers in Medicine. 2023 Nov 27;10.
39. Benjamens S, Dhunnoo P, Meskó B. The state of artificial intelligence-based FDA-approved medical devices and algorithms: an online database. npj Digital Medicine. 2020 Sep 11;3(1):118.
40. Kondrashova R, Klimeš F, Kaireit TF, May K, Barkhausen J, Stiebeler S, et al. Comparison of AI software tools for automated detection, quantification and categorization of pulmonary nodules in the HANSE LCS trial. Scientific Reports. 2024 Nov 13;14(1):27809.
41. Strubchevska O, Kozyk M, Kozyk A, Strubchevska K. The Role of Artificial Intelligence in Diagnostic Radiology. Cureus. 2024 Oct;16(10): e72173.