Artificial intelligence (AI) has emerged as a transformative trend in modern healthcare, offering advanced tools for diagnosis, clinical decision-making, and workflow optimization. This systematic review explores the integration of AI in three key medical domains: radiology, emergency medicine, and bone age estimation by using dental imaging. In radiology, AI has demonstrated diagnostic accuracies often exceeding human performance, especially in high-volume scenarios like the coronavirus disease 2019 (COVID-19) pandemic. In emergency departments, machine learning algorithms are enhancing triage, predicting patient outcomes, and reducing diagnostic delays. Similarly, AI-driven models like AlexNet, DASNet, and ResNet have significantly improved the precision of age estimation from orthopantomograms, outperforming traditional methods. Despite these advancements, the integration of AI is challenged by issues including algorithmic bias, data privacy concerns, and clinical workflow misalignment. The review highlights both the potential and the challenges of AI in healthcare, emphasizing the need for diverse data, transparency, and interdisciplinary collaboration.

Artificial intelligence, Radiology, Emergency medicine, Bone age estimation, Dental imaging, Orthopantomogram, Diagnostic accuracy, Healthcare technology integration.

Artificial intelligence (AI) is transforming modern healthcare. It provides effective tools for clinical decision support, predictive analytics, and data-driven diagnostics. The origins of it are in rule-based systems from the 1950s, and starting in the 1990s, machine learning (ML) and deep learning (DL) increased its usefulness. These computational advances have enabled machines to detect complex patterns, analyze massive datasets, and execute predictive tasks with minimal human input. As a result, clinical practice has been reshaped across several domains.[1-4]

Key healthcare domains where AI has shown disruptive potential include radiology, emergency medicine, and bone age assessment. AI has developed as a key component of image interpretation in radiology, increasing diagnostic speed, sensitivity, and specificity. Convolutional neural networks (CNNs) are deep learning models that have surpassed traditional image evaluation by automating diagnostic operations and detecting subtle errors.[5-8]

This change was highlighted by the COVID-19 pandemic. AI technologies assisted radiologists in managing increasing imaging volumes by selecting critical patients and assisting with diagnostic decisions, specifically with computed tomography (CT) and chest X-rays.[9-12]

AI helps in the emergency department (ED) to solve long-standing structural issues like overcrowding, a lack of clinicians, and inconsistent decision-making. These days, real-time AI-powered triage systems help with patient outcome prediction. It also provides resource allocation optimization and diagnostic delay minimization.[13-14]

Bone age assessment, specifically using dental radiography, is another interesting use. Conventional techniques like Demirjian’s or Nolla’s have limited accuracy and observer variability. AI provides reliable and accurate alternatives, frequently predicting ages within ±1 year of actual ages. Predictive performance has greatly increased with the use of models like AlexNet, ResNet, and DASNet in dental orthopantomogram (OPG) analysis.[15]

This systematic study assesses the importance of AI in healthcare with a focus on radiology, emergency medicine, and bone age assessment by using dental imaging. It gathers research on AI-driven models’ diagnostic performance, methodological advances, and therapeutic value. It also addresses ethical, technological, and practical issues, along with future research paths for responsible AI incorporation in clinical practice.[16]

Radiology has seen an immense revolution because of the incorporation of AI, specifically deep learning models. CNNs like ResNet and VGGNet have been trained on immense medical image datasets to detect anomalies with higher precision than human radiologists. These models work by learning pixel-level patterns from thousands of annotated scans. It allows them to detect subtle or early-stage disease characteristics.[17]

Several studies used these models on chest X-rays and CT scans, especially during the COVID-19 pandemic, when AI technologies were used to triage patients, emphasize severe cases, and help overburdened radiology departments. AI-enabled solutions enhanced diagnostic turnaround time and reduced the number of missed anomalies, providing a scalable solution for handling massive image volumes.[18]

Performance criteria often reported in these trials were sensitivity, specificity, and area under the curve (AUC), which frequently exceeded 0.90 for common diseases like pneumonia or pulmonary embolism. These findings not only demonstrate AI’s technological effectiveness but also indicate a paradigm change in radiological practice from manual image reading to enhanced diagnostic procedures.[19]

In the high-stress environment of the ED, AI has emerged as a key support system for triage, diagnosis, and prognosis. Researchers evaluated AI applications in the ED by using data from electronic health records, vital signs, and diagnostic imaging. Machine learning methods ranging from logistic regression to deep learning networks have been used to create prediction models of disease severity, hospital admissions, and mortality risk.[20]

Some models predicted sepsis with an AUC of 0.92 and had good sensitivity and specificity for diagnoses like appendicitis. Real-time triage systems trained on thousands of ED patients increased urgency evaluations by 25% and lowered patient wait times by approximately 19 minutes across institutions. Such models were validated by retrospective, prospective, and even randomized controlled trials.  Their performance was assessed by using pooled metrics like AUC, sensitivity, and specificity.[21]

Importantly, studies frequently used techniques like the quality assessment of diagnostic accuracy studies-2 (QUADAS-2) framework to address diagnostic bias and the grading of recommendations assessment, development and evaluation (GRADE) system to evaluate evidence quality, demonstrating a thorough and standardised approach to AI evaluation. Despite variances in model design and data inputs, the research generally agrees that AI improves ED workflow efficiency and clinical accuracy.[22]

Bone age estimation (BAE), specifically by using OPGs, has moved away from manual chart-based approaches towards AI-driven analysis. Traditional procedures like the Demirjian, Nolla, and Haavikko methods have limits in terms of accuracy and standardisation. For example, the Demirjian approach exaggerated age in men and women by 53.1% and 44.2%, respectively. Similarly, Nolla’s approach had an accuracy of 69% for males and 61.4% for females, indicating an underestimation.[23]

Modern experiments that used deep learning models like AlexNet, ResNet, and DASNet revealed considerably higher results. These CNNs were trained by using selected OPG datasets that eliminated images with artifacts, dental abnormalities, or low contrast. Preprocessing procedures like region of interest (ROI) segmentation and contrast enhancement were used to increase model input fidelity. The resulting models obtained a prediction accuracy of up to 97% within ±0.8 years of actual age.[24]

Some research integrated CNNs with machine learning classifiers like support vector machines and decision trees to create hybrid models with even higher resilience. These ensemble techniques achieved a compromise between feature extraction and classification strength while being less prone to overfitting. AI accurately detected anatomical signals like apical foramen development and pulp chamber alterations, even when physical examination would be inconclusive or time-consuming.[25]

Furthermore, advances like multilayer perceptrons and transformer-CNN hybrids enhanced prediction under changeable imaging settings, demonstrating AI’s flexibility to a wide range of forensic and clinical scenarios. These findings support the notion that dental features, due to their resilience to decay and environmental deterioration, are ideal for AI-based age estimation.[26,27]

A dual-method approach was employed to evaluate the effectiveness and applications of AI in healthcare. This review specifically focuses on age estimation through panoramic radiographs (orthopantomograms, or OPGs) and diagnostic decision-making in ED settings.

Literature review and data sources
Comprehensive literature searches were performed across databases including PubMed, Embase, Web of Science, Cochrane Library, IEEE Xplore, and Scopus. Studies were selected based on inclusion criteria that required human subjects, clinical relevance, and use of AI for diagnosis, triage, or outcome prediction. Opinion articles and studies lacking real-world clinical application were excluded. Reference lists of relevant articles were also manually reviewed to ensure completeness.

Eligibility and selection
Eligibility was determined independently by two reviewers by using the Covidence software. For the OPG studies, only high-quality radiographs (free of artifacts, anomalies, or fixed appliances) were included. For the ED-based studies, randomized controlled trials, cohort, and cross-sectional designs were accepted if they assessed clinical outcomes or decision-making processes enhanced by AI.

Data extraction and evaluation
Data from selected studies were extracted using standardized Excel-based forms. For the OPG-focused research, both traditional methods (Nolla, Demirjian, Willem, Haavikko) and AI models (e.g., CNNs like AlexNet and ResNet) were evaluated. Preprocessing of images involved contrast enhancement and segmentation. Model performance was assessed using metrics such as accuracy (±1 year), mean absolute error (MAE), and computational efficiency.

In ED-related studies, extracted variables included study and participant characteristics, AI tools used, training/validation methods, diagnostic or triage performance (AUC, sensitivity, specificity), and ethical concerns. Risk of bias was evaluated using QUADAS-2 and the Cochrane risk of bias tool.

Synthesis and analysis
Due to heterogeneity, narrative synthesis was employed for both domains. Where applicable, meta-analyses with random-effects models and 95% confidence intervals were conducted. Heterogeneity was evaluated via the I² statistic, and subgroup analyses were based on AI types and clinical applications. Reporting bias was assessed using Egger’s test, and the certainty of evidence was graded using the GRADE approach.

Advancing age prediction accuracy with AI and orthopantomograms
The study by Ayman et al. demonstrates that AI-based age estimation methods using OPGs significantly outperform traditional dental assessment techniques in terms of accuracy and precision. Traditional methods like Demirjian, Nolla, Willems, and Haavikko show moderate accuracy with inherent limitations such as over- or underestimation and poor generalizability. AI models, particularly those using CNNs, ensemble approaches, and hybrid architectures, achieve up to 97% accuracy within a ±0.8-year range. This highlights the transformative potential of AI in providing highly accurate, generalizable, and real-time forensic age predictions, especially when enhanced with techniques like adaptive image preprocessing.[28]

Impact of AI on diagnosis, triage, and outcome prediction in EDs
The study by Adrita et al. highlights that AI models significantly outperform traditional methods in ED settings across key performance metrics. An analysis of 29 studies from 1,245 screened records showed that AI demonstrated superior diagnostic accuracy, with a pooled AUC of 0.88 (95% confidence interval (CI) 0.85-0.91). This indicates reliable and precise identification of clinical conditions. In terms of triage efficiency, AI systems reduced average patient wait times by 18.7 minutes (95% CI: 12.4-25.0), suggesting notable improvements in workflow and patient throughput. Additionally, AI models achieved a pooled sensitivity of 0.85 (95% CI: 0.81–0.89) for predicting hospital admissions, reflecting strong potential in outcome forecasting and decision support. These results collectively underscore the value of AI in enhancing ED decision-making processes. However, the study also emphasizes that despite these advancements, real-world implementation remains challenged by ethical concerns, data transparency, safety, and the need for large-scale validation. Continued research is necessary to ensure effective, equitable, and safe integration of AI into clinical practice.[29]

Sustainable radiology: Advancing toward net-zero with AI and climate resilience
The review by Adeponle et al., radiology is evolving to meet the United Nations Sustainable Development Goals by adopting environmentally responsible practices. With high energy demands and the use of hazardous materials, the field faces growing pressure to reduce its ecological footprint. Sustainable strategies include energy-efficient imaging, improved waste management, and the use of eco-friendly technologies, all aimed at delivering equitable and high-quality care.

Climate change also impacts radiology by affecting equipment performance and increasing demand for imaging due to emerging climate-related diseases. To maintain diagnostic precision and operational continuity, radiology departments must invest in climate-resilient infrastructure.

AI is playing a critical role in this transition. AI optimizes scan protocols, reduces unnecessary imaging, enhances workflow efficiency, and supports remote diagnostics, collectively lowering energy use and carbon emissions. Predictive maintenance also extends equipment lifespan, reducing waste.

However, the ethical use of AI in radiology is essential. Safeguarding patient privacy, ensuring informed consent, and addressing algorithmic bias are crucial to maintaining trust and fairness. AI must complement, not replace, clinical judgment.

While AI and predictive analytics are driving operational and diagnostic improvements, challenges remain in workforce capacity and uniform implementation. Ongoing innovation and ethical oversight are key to a sustainable and effective future for radiology.[30]

Table 1: Summary of AI applications in key healthcare domains

Challenges and ethical considerations in AI integration
While AI has demonstrated exceptional findings across various domains of healthcare, integration is not without significant challenges (Table 1). These limitations include technical, ethical, operational, and systemic dimensions. It must be addressed thoughtfully to ensure AI systems are safe, fair, and sustainable in clinical practice.

Recent breakthroughs in radiology, one of the most technically sophisticated sectors for AI deployment, have been linked with wider environmental goals like reaching net-zero carbon emissions. AI contributes favourably by reducing redundant imaging, optimising energy use, and selecting the best imaging modalities based on past data. Furthermore, AI-enabled predictive maintenance increases the life of imaging equipment, minimising waste due to hardware replacement. Remote diagnostics help minimise patient travel, which adds to sustainability. Despite these environmental benefits, ethical considerations remain important to AI’s application in radiology.

Patient privacy and data security are critical, as AI systems rely on massive amounts of sensitive personal health data. Data must be completely anonymized and safeguarded against breaches, which is sometimes challenging due to the decentralized nature of healthcare data repositories. Furthermore, informed consent is frequently overlooked, as patients are not always informed of how their data is being used or by whom, particularly when AI algorithms operate in the background of diagnostic processes.

Training AI models on non-diverse or imbalanced datasets might worsen algorithmic bias, which causes biased diagnostic results in radiology. Systematic testing on a variety of datasets and open reporting of AI system performance across demographics are required to overcome this. AI should support expert clinical judgement rather than take the role of radiologists. To establish responsibility, maintain the clinician-patient relationship, and confirm AI suggestions, human monitoring is essential. Radiologists must participate in algorithm development and decision-making processes as AI takes on more administrative and diagnostic tasks. Obstacles related to the radiology workforce include ageing practitioners, a lack of staff, and uneven AI adoption throughout institutions. To guarantee a safe and successful integration of AI into routine practice, training and standards must be invested in.

Dental radiography is used for bone age estimation, but there are technical and adoption issues. AI models like ResNet, DASNet, and AlexNet have superior accuracy, but deep neural networks have higher mean error rates than regression-based methods. Additionally, AI models are often developed without clinical training, leading to systems that do not align with real-world clinical workflows or ethical expectations. This disconnect can lead to reduced trust and low adoption among healthcare professionals like dental practitioners, who express reluctance to embrace AI due to system complexity, high implementation costs, and erosion of interpersonal patient care.

This focused review highlights the transformative role of AI across three critical domains in healthcare: radiology, emergency medicine, and dental-based bone age estimation. In each area, AI has demonstrated measurable improvements in diagnostic precision, workflow efficiency, and clinical decision-making. AI has suggested its growing utility in augmenting healthcare delivery.

AI’s application in dental age estimation, particularly using OPGs, represents another significant advance. Traditional methods like those by Demirjian and Nolla are limited by inter-observer variability and population-specific biases. AI-driven models, especially CNN-based architectures, have achieved accuracies within ±0.8 years, improving reliability in forensic and pediatric applications. Preprocessing techniques like contrast enhancement and segmentation further enhance model performance. Nonetheless, these systems must be seamlessly integrated into existing dental workflows and gain clinician trust through improved usability and interpretability.[28]

In ED, AI models are increasingly being deployed for triage optimization and outcome prediction. Studies demonstrate that these systems reduce patient wait times and improve prioritization, with pooled AUC values nearing 0.88. Such tools can support overstretched clinical teams by enhancing throughput and patient safety. However, the deployment of AI in high-stakes ED environments requires rigorous validation, real-time adaptability, and continuous clinician oversight to ensure reliability and ethical use.[29]

In radiology, deep learning models, particularly CNNs, have shown exceptional performance in interpreting imaging modalities like CT scans and X-rays. These systems automate routine diagnostic tasks, reduce radiologist workload, and prove particularly valuable during periods of high clinical demand, such as the COVID-19 pandemic. Reported metrics, including sensitivity, specificity, and AUC, often exceed 0.90. It indicates diagnostic accuracy on par with or surpassing that of human experts. Despite these advances, clinical adoption is hindered by concerns around data privacy, lack of algorithm transparency, and limited interpretability of outputs.[30]

While the specific implementations vary, common barriers to widespread AI integration span all domains. These include ethical concerns (e.g., privacy, consent), algorithmic bias arising from non-representative training data, and variability in data standards across institutions. Clinician resistance often stems from the complexity of AI systems, lack of transparency in decision-making, and perceived threats to professional autonomy. Overcoming these challenges requires interdisciplinary collaboration, regulatory oversight, and clinician-centered AI design. Efforts must prioritize explainability, fairness, and robust real-world validation to foster safe and equitable adoption.

AI should be viewed not as a replacement for clinical judgment but as a complementary tool that supports and enhances human expertise. Moving forward, fostering trust, ensuring equitable outcomes, and building infrastructure for continuous learning and feedback will be key to unlocking the full potential of AI in healthcare.

By improving diagnostic accuracy, speeding up clinical procedures, and resolving systemic inefficiencies, AI is completely changing the way healthcare is delivered. The clinical value is evident in its uses in radiography, hospital ED, and dental bone age estimation. However, ethical protections, fair training datasets, and clinician integration are necessary for AI to realize its full potential. AI should be used to supplement human knowledge, not to replace it. The safe and successful integration of AI in routine clinical practice will depend heavily on multidisciplinary cooperation, regulatory supervision, and targeted clinician education.

Not reported

No funding

Corresponding Author:
Samatha Ampeti, PhD
Department of Pharmacology
Kakatiya University, University College of Pharmaceutical Sciences, Warangal, TS, India
Email: ampetisamatha9@gmail.com

Co-Authors:
Patel Nirali Kirankumar, Mansi Srivastava, Raziya Begum Sheikh, Shubham Ravindra Sali
Independent Researcher, Department of Content
medtigo India Pvt Ltd, Pune, India

All authors contributed to the conceptualization, investigation, and data curation by acquiring and critically reviewing the selected articles. They were collectively involved in the writing – original draft preparation and writing – review & editing to refine the manuscript. Additionally, all authors participated in the supervision of the work, ensuring accuracy and completeness. The final manuscript was approved by all named authors for submission to the journal.

Not applicable

None

None

https://doi.org/10.63096/medtigo3092234

Patel NK, Samatha A, Mansi S, Raziya BS, Shubham RS. The Role of AI in Healthcare: A Focused Review on Radiology, Emergency Department and Dental Age Estimation. medtigo J Emerg Med. 2025;2(3):e3092234. doi:10.63096/medtigo3092234 Crossref