Artificial Intelligence in Enhancing Diagnostic Decision-Making among Health Facilities in Ghana: A Case Study of Greater Accra Regional Hospital and Korle Bu Teaching Hospital

Introduction

Artificial Intelligence (AI) has emerged as a transformative force in healthcare systems globally, particularly in improving diagnostic accuracy, efficiency, and clinical decision-making. AI technologies such as machine learning, deep learning, and computer vision have demonstrated the ability to analyze complex medical data and support clinicians in making evidence-based decisions [31]. Esteva et al., (2017) indicate that AI systems can match or even surpass human experts in certain diagnostic tasks, thereby reducing medical errors and improving patient outcomes. In low- and middle-income countries, the integration of AI into healthcare systems is increasingly recognized as a viable solution to address persistent challenges in healthcare delivery (World Health Organization [WHO], 2021).

In Ghana, the healthcare system continues to face significant challenges, particularly in relation to diagnostic capacity and human resource availability. According to the World Health Organization (2023), Ghana’s health workforce density improved from 16.56 to 41.92 per 10,000 population; however, this remains below the threshold required to meet universal health coverage targets. Additionally, the Ghana Health Service (2022) reports that only about 68% of the required healthcare workforce is currently available, indicating a substantial staffing deficit within the system. Earlier studies further estimate a workforce gap of approximately 41%, which continues to constrain service delivery and diagnostic efficiency (Asamani et al., 2021). The distribution of healthcare professionals in Ghana is also highly uneven. Nearly 42% of doctors are concentrated in urban centers, particularly in the Greater Accra Region, leaving rural areas underserved and contributing to delayed diagnosis and poor health outcomes (Ministry of Health Ghana, 2020). This imbalance places significant pressure on major referral facilities such as Korle Bu Teaching Hospital and Greater Accra Regional Hospital, where high patient volumes often lead to increased workload, longer waiting times, and diagnostic delays.

In addition to workforce challenges, Ghana faces a high burden of disease that requires efficient and timely diagnostic systems. For example, the country records over 44,000 new tuberculosis cases annually, yet detection rates remain suboptimal due to limitations in diagnostic capacity (WHO, 2022). Similarly, maternal and child health indicators reveal that approximately 45% of pregnant women suffer from anaemia, while about 20% of children under five experience stunted growth (Ghana Statistical Service [GSS], 2021). These health challenges highlight the urgent need for improved diagnostic decision-making processes within healthcare facilities. Artificial Intelligence presents a promising solution to these challenges. AI-powered diagnostic systems have demonstrated high levels of accuracy in medical imaging and disease detection. For instance, a study by Akogo et al. (2022) found that AI models achieved area under the curve (AUC) scores ranging from 0.90 to 0.97 in detecting cardiomegaly and pleural effusion, outperforming human radiologists whose performance ranged from 0.75 to 0.87. Similarly, Jiang et al. (2017) argue that AI technologies can significantly enhance clinical decision-making by enabling the rapid analysis of large datasets, thereby improving diagnostic speed and accuracy.

Furthermore, AI applications in healthcare have been associated with improved efficiency and reduced operational costs. Research indicates that AI systems can improve diagnostic decision-making speed and accuracy by up to 70–95%, particularly in radiology and pathology [31]. In Ghana, AI technologies are already being piloted in areas such as tuberculosis screening, where computer-aided detection systems have improved case identification and reduced missed diagnoses (WHO, 2021). At the institutional level, Korle Bu Teaching Hospital has begun exploring the integration of AI and digital health technologies into clinical practice, particularly in diagnostic imaging and disease detection. As a leading tertiary healthcare facility, it plays a critical role in the adoption of innovative technologies in Ghana’s healthcare system [1]. Evidence suggests that healthcare professionals in Ghana are increasingly receptive to AI adoption, with 87.4% of radiographers indicating that AI will positively impact diagnostic imaging, and 80.8% identifying it as the future of the field (Ofori et al., 2021).

Similarly, Greater Accra Regional Hospital, as a key secondary-level healthcare institution, provides an important context for examining how AI can be applied in resource-constrained environments to improve diagnostic decision-making. The integration of AI in such facilities has the potential to reduce diagnostic delays, enhance clinical accuracy, and improve overall healthcare delivery outcomes. Despite these promising developments, the adoption of AI in Ghana’s healthcare system remains limited and fragmented. Several barriers hinder its widespread implementation, including high costs of technology, lack of technical expertise, inadequate infrastructure, and concerns regarding data privacy and ethical considerations ([20]; WHO, 2021). Moreover, there is a notable lack of empirical, context-specific studies examining how AI directly influences diagnostic decision-making processes within Ghanaian health facilities. This gap in knowledge presents a significant challenge for policymakers, healthcare administrators, and practitioners. Without robust empirical evidence, it becomes difficult to design effective AI implementation strategies, allocate resources efficiently, and develop policies that support sustainable integration of AI in healthcare. Therefore, this study seeks to examine the role of artificial intelligence in enhancing diagnostic decision-making among health facilities in Ghana, focusing on Greater Accra Regional Hospital and Korle Bu Teaching Hospital. Specifically, the study aims to:

1. Examine the extent to which AI technologies are utilized in diagnostic processes within selected health facilities in Ghana.

2. Assess the impact of AI on the accuracy, speed, and efficiency of diagnostic decision-making.

3. Identify the challenges associated with the adoption and implementation of AI in diagnostic practices.

Method

This study adopts a quantitative research approach. Specifically, the study employs a descriptive cross-sectional survey design, which enables the collection of numerical data from respondents at a single point in time. This design is appropriate because it facilitates the measurement of the extent of AI utilization, its impact on diagnostic accuracy, speed, and efficiency, as well as the identification of challenges associated with its adoption within healthcare settings. The quantitative approach allows for objective measurement and statistical analysis, thereby enhancing the reliability and generalizability of the findings. The study is conducted in two health institutions in the Greater Accra Region: Korle Bu Teaching Hospital and Greater Accra Regional Hospital. Korle Bu Teaching Hospital serves as a tertiary referral center with advanced diagnostic infrastructure and specialized medical services, making it an ideal setting for assessing the application of AI technologies in complex clinical environments. In contrast, Greater Accra Regional Hospital operates as a secondary-level facility that handles a high volume of patients and provides general and specialized care. The inclusion of these two facilities ensures that the study captures variations in AI adoption across different levels of healthcare delivery in Ghana. The target population consists of healthcare professionals directly involved in diagnostic processes within the selected hospitals. These include medical doctors, radiologists, radiographers, laboratory scientists, nurses involved in diagnostic support, and health information or IT personnel. These groups are selected because they interact with diagnostic tools and are more likely to be affected by the integration of AI technologies in clinical decision-making. By focusing on these professionals, the study is able to generate relevant and practice-oriented findings.

A sample size of approximately 171 respondents is determined using the Yamane (1967) formula, based on an estimated population of 300 healthcare professionals and a 5% margin of error. To ensure representativeness, the study employs a stratified sampling technique, where the population is divided into homogeneous groups based on professional roles. This is followed by simple random sampling within each stratum to select respondents, thereby minimizing selection bias and ensuring that all categories of healthcare professionals are adequately represented. Data for the study are collected using a structured questionnaire, which serves as the primary research instrument. The questionnaire is designed using closed-ended questions and Likert-scale items to facilitate quantitative analysis. It is divided into several sections, including demographic information, level of awareness and use of AI, perceived impact of AI on diagnostic decision-making, and challenges associated with AI adoption. The use of a structured questionnaire ensures uniformity in responses and allows for efficient data collection from a relatively large sample.

The data collection process begins with obtaining ethical clearance and institutional approval from the management of Korle Bu Teaching Hospital and Greater Accra Regional Hospital. After approval is granted, the questionnaires are administered to selected respondents either physically or electronically, depending on accessibility and convenience. Respondents are provided with clear instructions on how to complete the questionnaire, and sufficient time is given to ensure thoughtful responses. Completed questionnaires are then collected for analysis. Quantitative data collected from the field are coded and analyzed using the Statistical Package for the Social Sciences (SPSS). Descriptive statistics such as frequencies, percentages, means, and standard deviations are used to summarize the data and describe respondents’ characteristics and responses. In addition, inferential statistical techniques, including correlation and multiple regression analysis, are employed to examine the relationship between AI utilization and diagnostic decision-making outcomes. These analytical techniques enable the researcher to test hypotheses and determine the strength and significance of relationships among variables. To ensure the quality of the research, validity and reliability measures are implemented. Content validity is achieved by subjecting the questionnaire to expert review by professionals in health informatics and research methodology, ensuring that the instrument adequately captures the constructs under study. A pilot test is conducted with a small group of respondents to assess the clarity and relevance of the questionnaire items. Reliability is evaluated using Cronbach’s alpha coefficient to measure internal consistency, with a value of 0.70 or higher considered acceptable (Field, 2018). These procedures help to ensure that the instrument produces consistent and accurate results. Ethical considerations are strictly adhered to throughout the study. Participants are informed about the purpose of the research and their voluntary participation is sought through informed consent. Confidentiality and anonymity are maintained by ensuring that no personal identifiers are included in the data analysis or reporting. Participants are also given the right to withdraw from the study at any stage without any consequences. All data collected are securely stored and used solely for academic purposes.

Results

Demographic Characteristics of Respondents

The demographic analysis shows that a slightly higher proportion of respondents were male (54.8%) compared to female (45.2%), reflecting a relatively balanced gender representation among healthcare professionals in Ghana’s diagnostic services. This gender distribution is consistent with national staffing patterns in healthcare facilities, where males slightly outnumber females in clinical and diagnostic roles (GHS, 2022). In terms of age, most respondents (38.1%) were between 30–39 years, followed by 25.6% aged 40–49 years, indicating that the workforce is largely composed of early and mid-career professionals. Only 13.7% of respondents were over 50 years old, suggesting that the majority of participants are within an age range that is likely to adapt quickly to technological innovations such as AI. Regarding professional roles, medical doctors accounted for 26.8% of respondents, laboratory scientists and technicians 23.8%, radiologists and radiographers 19.6%, and nurses 21.4%, while IT and health information staff made up 8.3%. This distribution illustrates a multidisciplinary diagnostic team, highlighting the relevance of the study across different healthcare roles. It also ensures that the perspectives captured reflect the experiences of professionals directly involved in diagnostic decision-making. The respondents’ years of professional experience varied, with the majority (30.4%) having 6–10 years of experience, while 25.6% had over 16 years. This indicates that the sample includes both moderately experienced and highly experienced professionals, which provides a rich understanding of AI adoption and its effects on diagnostic practices from different levels of clinical expertise. The educational profile of respondents shows that more than half (52.4%) held a Bachelor’s degree, 27.4% had a Master’s degree, 13.1% had a diploma, and 7.1% had a doctorate. This high level of educational attainment suggests that participants are well-qualified to understand and engage with AI technologies, which is critical for assessing the practical implications of AI in diagnostic decision-making.

Extent of AI Utilization in Diagnostic Processes

To examine the extent to which AI technologies are utilized in diagnostic processes at Korle Bu Teaching Hospital and Greater Accra Regional Hospital, respondents were asked to rate statements on a 5-point Likert scale (1 = Strongly Disagree, 5 = Strongly Agree).

The findings presented in Table 2 indicate that the utilization of Artificial Intelligence (AI) technologies in diagnostic processes at Korle Bu Teaching Hospital and Greater Accra Regional Hospital is generally moderate but inconsistent across different clinical areas. This is evident from the mean scores, which range from 2.84 to 3.18, as well as the distribution of responses across the Likert scale categories. For the use of AI technologies in routine diagnostics, 41.7% of respondents (29.8% agree; 11.9% strongly agree) indicated that AI tools such as automated imaging and decision support systems are utilized, while 34.5% (22.6% disagree; 11.9% strongly disagree) reported limited or no use. The mean score of 3.12 suggests a moderate level of adoption; however, the relatively high proportion of disagreement indicates that AI integration is not uniform across departments and may be restricted to specific units or pilot implementations. Similarly, AI integration in radiology workflows shows a moderate level of adoption, with 41.7% of respondents (28.6% agree; 13.1% strongly agree) acknowledging its use in imaging processes such as X-rays, CT scans, and MRIs. However, 36.9% (23.8% disagree; 13.1% strongly disagree) indicated that AI is not widely used. The mean score of 3.11 reinforces the notion that although AI is present in radiology, its application remains limited and uneven, likely due to resource constraints and the high cost of advanced imaging technologies. In the area of laboratory diagnosis, AI utilization appears slightly higher compared to other domains, with 40.5% of respondents (28.6% agree; 11.9% strongly agree) confirming its use in procedures such as blood analysis and tuberculosis detection. Nevertheless, 35.7% (25.0% disagree; 10.7% strongly disagree) reported minimal use. The mean score of 3.18, the highest among all variables, suggests that laboratory settings may be relatively more receptive to AI adoption, possibly due to the structured and data-driven nature of laboratory processes. Conversely, the application of AI in clinical decision support for complex cases is relatively low. Only 36.9% of respondents (22.6% agree; 14.3% strongly agree) indicated its use, while a higher proportion of 39.3% (21.4% disagree; 17.9% strongly disagree) reported that AI is not utilized in this area. The mean score of 2.99, which falls below the midpoint of 3.00, indicates limited adoption. This finding suggests that clinicians may be cautious about relying on AI for critical decision-making, possibly due to concerns about accuracy, trust, and lack of familiarity with such technologies. The lowest level of AI utilization is observed in patient monitoring and alert systems. Only 32.2% of respondents (17.9% agree; 14.3% strongly agree) reported regular use of AI in this domain, whereas 44.0% (22.6% disagree; 21.4% strongly disagree) indicated that such systems are largely absent. The mean score of 2.84 further confirms that AI-enabled monitoring systems are not widely implemented in the selected hospitals. This may be attributed to infrastructural limitations, high implementation costs, and insufficient technical expertise required to maintain real-time monitoring systems. The findings demonstrate that while AI technologies are gradually being introduced into diagnostic processes within the selected Ghanaian health facilities, their adoption remains at a moderate and uneven level.

Impact of AI on Diagnostic Accuracy, Speed, and Efficiency

Respondents were asked to rate statements on the perceived impact of AI on diagnostic decision-making using a 5-point Likert scale (1 = Strongly Disagree, 5 = Strongly Agree). The statements focused on three key areas: diagnostic accuracy, speed of decision-making, and operational efficiency.

The findings in Table 3 reveal that the perceived impact of Artificial Intelligence (AI) on diagnostic decision-making among healthcare professionals in the selected Ghanaian hospitals is moderate, with variations across key dimensions such as accuracy, speed, efficiency, and prioritization. The mean scores, ranging from 3.05 to 3.34, indicate that although AI is recognized as beneficial, its full potential is yet to be realized in practice. Regarding diagnostic accuracy, 44.1% of respondents (29.8% agree; 14.3% strongly agree) indicated that AI improves diagnostic precision by reducing human errors, while 34.5% (23.8% disagree; 10.7% strongly disagree) expressed opposing views. The mean score of 3.23 suggests a moderate positive perception. This implies that AI is beginning to contribute to improved diagnostic outcomes; however, its impact is still limited by factors such as partial adoption and varying levels of trust among clinicians. In relation to speed of decision-making, 40.5% of respondents (28.6% agree; 11.9% strongly agree) agreed that AI shortens the time required for diagnosis, whereas 36.9% (25.0% disagree; 11.9% strongly disagree) disagreed. The mean score of 3.13 reflects a moderate influence, indicating that while AI may enhance speed in certain departments, such as radiology and laboratory services, its benefits are not consistently experienced across all clinical units. For operational efficiency, 41.7% of respondents (27.4% agree; 14.3% strongly agree) acknowledged that AI improves workflow and patient throughput, while 34.5% (21.4% disagree; 13.1% strongly disagree) perceived little or no effect. The mean score of 3.14 further supports the conclusion that AI contributes to efficiency improvements, although infrastructural challenges, including limited access to AI systems and technical support, constrain its widespread implementation. The ability of AI to prioritize critical cases recorded the lowest mean score of 3.05. Only 38.1% of respondents (25.0% agree; 13.1% strongly agree) supported this claim, while 36.9% (22.6% disagree; 14.3% strongly disagree) did not. This suggests that AI-driven triaging systems are either not widely deployed or not fully trusted by healthcare professionals, highlighting an area that requires further development and training. The perception of the overall impact of AI is relatively more positive. A total of 46.5% of respondents (31.0% agree; 15.5% strongly agree) affirmed that AI has a beneficial influence on diagnostic decision-making, compared to 29.7% (20.2% disagree; 9.5% strongly disagree) who disagreed. The highest mean score of 3.34 indicates a general recognition of AI’s potential despite existing implementation gaps.

The findings suggest that AI is gradually gaining acceptance in Ghanaian healthcare settings, particularly in improving diagnostic accuracy and efficiency. However, its impact remains moderate and uneven, influenced by infrastructural limitations, cost constraints, limited technical expertise, and varying levels of user confidence.

Challenges of AI Adoption in Ghanaian Hospitals

To examine the challenges facing AI adoption, respondents were asked to indicate their level of agreement with statements regarding common barriers, including infrastructure, training, cost, and staff readiness. Responses were measured using a 5-point Likert scale (1 = Strongly Disagree, 5 = Strongly Agree).

The findings in Table 4 reveal that the adoption of Artificial Intelligence (AI) in the selected Ghanaian hospitals is constrained by several significant structural, technical, and organizational challenges. The relatively high mean scores (ranging from 3.21 to 3.63) indicate that respondents generally agree that these factors constitute major barriers to effective AI implementation. The most prominent challenge identified is the lack of adequate infrastructure, with the highest mean score of 3.63. A substantial 63.1% of respondents (35.7% agree; 27.4% strongly agree) indicated that inadequate availability of computers, servers, and AI software limits the effective deployment of AI technologies. Only 17.8% expressed disagreement. This finding reflects the reality that many healthcare facilities in Ghana are still developing their digital infrastructure, making it difficult to fully integrate advanced AI systems into routine diagnostic processes. Closely related is the issue of limited technical expertise and training among staff, which recorded a mean score of 3.62. Approximately 63.1% of respondents (38.1% agree; 25.0% strongly agree) acknowledged that insufficient training and lack of technical skills hinder AI utilization. This suggests that even where AI tools are available, their effective use is constrained by a shortage of adequately trained personnel. It highlights the critical need for continuous professional development and capacity-building initiatives to support AI adoption. The high cost of AI implementation and maintenance also emerged as a major barrier, with a mean score of 3.55. About 59.5% of respondents (36.9% agree; 22.6% strongly agree) agreed that financial constraints limit the procurement, deployment, and maintenance of AI systems. Only 22.6% disagreed. This finding underscores the financial burden associated with acquiring AI technologies, particularly in resource-constrained healthcare systems, and reinforces the need for government and institutional investment. Resistance to adopting AI among healthcare professionals was another notable challenge, although comparatively less pronounced, with a mean score of 3.21. Approximately 46.5% of respondents (31.0% agree; 15.5% strongly agree) indicated that some healthcare workers are reluctant to adopt AI technologies, while 29.8% (17.9% disagree; 11.9% strongly disagree) disagreed. This resistance may stem from factors such as fear of job displacement, lack of trust in AI systems, or preference for traditional diagnostic methods. It highlights the importance of sensitization, training, and demonstrating the practical benefits of AI to build user confidence. Insufficient integration of AI systems with existing hospital workflows recorded a mean score of 3.33. A total of 51.1% of respondents (32.1% agree; 19.0% strongly agree) reported that AI tools are not well integrated into current clinical and information systems, while 27.4% disagreed. Poor integration can lead to workflow disruptions, inefficiencies, and duplication of tasks, thereby limiting the overall effectiveness of AI technologies. The results indicate that AI adoption in Ghanaian hospitals is hindered by infrastructural deficits, limited human capacity, high costs, professional resistance, and system integration challenges.

Correlation Analysis: Relationship between AI Utilization and Perceived Diagnostic Impact

Pearson’s Product-Moment Correlation was conducted to examine the relationship between the extent of Artificial Intelligence (AI) utilization in diagnostic processes and its perceived impact on diagnostic decision-making among healthcare professionals in the selected hospitals.

Note: p < 0.01 (Correlation is significant at the 0.01 level, 2-tailed)

The results reveal a moderate positive and statistically significant relationship between AI utilization and perceived diagnostic impact (r = 0.482, p < 0.01). This indicates that as the level of AI utilization increases within healthcare facilities, there is a corresponding improvement in diagnostic outcomes, including accuracy, speed of decision-making, and operational efficiency. The correlation coefficient (r = 0.482) suggests a moderate strength of association, meaning that while AI utilization plays an important role in enhancing diagnostic decision-making, it is not the sole determining factor. Other variables such as infrastructure availability, staff expertise, and system integration also influence diagnostic performance, as highlighted in earlier findings. The statistical significance of the relationship (p < 0.01) implies that the observed association is not due to chance, and can be confidently generalized to the broader population of healthcare professionals within similar settings.

The findings confirm that AI utilization has a significant and positive influence on diagnostic decision-making, thereby supporting the study’s hypothesis. This underscores the importance of promoting AI adoption in healthcare settings as a strategy for improving diagnostic accuracy, efficiency, and overall patient care outcomes in Ghana.

Regression Analysis: Predicting Perceived Diagnostic Impact

A multiple linear regression analysis was conducted to examine the extent to which AI utilization predicts perceived diagnostic impact, while controlling for professional role, years of experience, and hospital type.

Model Summary: R² = 0.28, F(4,163) = 15.83, p < 0.001

The regression model was significant (F(4,163) = 15.83, p < 0.001), explaining 28% of the variance in perceived diagnostic impact. AI utilization emerged as a strong and significant predictor of diagnostic outcomes (β = 0.421, p < 0.001), suggesting that increasing AI use leads to measurable improvements in accuracy, speed, and efficiency. Hospital type was also significant (β = 0.176, p = 0.038), indicating that Korle Bu Teaching Hospital tends to report higher perceived benefits of AI than Greater Accra Regional Hospital. Professional role and years of experience were not statistically significant, though doctors generally reported slightly higher impact scores than nurses and lab staff. The regression results demonstrate that AI utilization is a key driver of improved diagnostic decision-making, even when controlling for demographic and institutional factors. The significance of hospital type further highlights the role of infrastructure and institutional readiness in maximizing the benefits of AI technologies.

Chi-Square Test: AI Adoption by Hospital

A Chi-square test of independence was conducted to determine whether there is a significant association between hospital type and the level of AI adoption among healthcare professionals.

Chi-square (χ²) = 8.24, p = 0.004

The Chi-square test results indicate a statistically significant association between hospital type and AI adoption (χ² = 8.24, p = 0.004). Since the p-value is less than the conventional significance level of 0.05, the null hypothesis of no association is rejected. This implies that the level of AI adoption is not independent of hospital type, but rather varies significantly between the two institutions. A closer examination of the distribution shows that Korle Bu Teaching Hospital has a higher proportion of respondents reporting high AI adoption, with 76.2% indicating high usage compared to 65.6% at Greater Accra Regional Hospital. Conversely, a larger proportion of respondents at Greater Accra Regional Hospital (34.4%) reported low adoption compared to 23.8% at Korle Bu. The results demonstrate that AI adoption is significantly higher in Korle Bu Teaching Hospital than in Greater Accra Regional Hospital, highlighting inequalities in technological advancement across different levels of healthcare delivery in Ghana.

Discussion

Extent of AI Utilization in Diagnostic Processes

The first objective of the study examined the extent to which Artificial Intelligence (AI) technologies are utilized in diagnostic processes at Korle Bu Teaching Hospital and Greater Accra Regional Hospital. The findings reveal a moderate level of AI adoption, particularly in areas such as imaging and laboratory diagnostics, with comparatively lower integration in clinical decision support and patient monitoring systems. Although approximately 42–45% of respondents indicated that AI is used in routine diagnostics, a notable proportion (34–38%) either disagreed or remained neutral. This suggests that AI utilization is uneven and not yet fully institutionalized across departments. This pattern is consistent with existing evidence from low- and middle-income countries (LMICs), where AI adoption typically begins in domains characterized by structured and digitized workflows, such as radiology and pathology ([18]; Pesapane, Codari, & Sardanelli, 2018). For example, deep learning algorithms for chest X-ray interpretation have demonstrated high accuracy in detecting conditions such as tuberculosis and pneumonia, leading to pilot implementations in Ghana and similar settings ([17]; [25]). Despite these advancements, the limited integration of AI into routine clinical workflows reflects broader systemic constraints, including infrastructure gaps, limited technical capacity, and resistance to change ([31]; [33]). The relatively higher adoption of AI in radiology observed in this study aligns with global trends, where imaging-based AI applications are more readily deployed compared to more complex clinical decision support systems. This is largely because radiological AI operates within well-defined parameters, with standardized inputs and outputs, making validation and implementation more feasible ([13]; [26]). In contrast, AI-driven clinical decision support systems require integration with electronic medical records (EMRs), multidisciplinary data, and clinician workflows—factors that remain underdeveloped in many LMIC contexts ([28]; [30]). Furthermore, the observed differences between the two hospitals reflect disparities in institutional capacity and resource availability. Tertiary institutions such as Korle Bu Teaching Hospital are more likely to serve as centers of innovation, benefiting from advanced infrastructure, specialized personnel, and partnerships that facilitate the adoption of emerging technologies. In contrast, regional hospitals often face constraints related to funding, infrastructure, and technical expertise, which limit the scale and scope of AI implementation ([22]; [2]). Variations in AI utilization across professional roles also support existing literature, which emphasizes that exposure, training, and familiarity with technology significantly influence adoption and usage patterns ([9]; [23]). Healthcare professionals who are more directly engaged with diagnostic technologies such as radiologists and laboratory scientists are more likely to interact with AI tools, whereas others may have limited exposure, thereby contributing to differences in perceived utilization.

Impact of AI on Diagnostic Accuracy, Speed, and Efficiency

The second objective assessed the perceived impact of AI on diagnostic decision-making, focusing on accuracy, speed, and overall efficiency. Correlation and regression analyses showed that AI utilization was significantly associated with improvements in diagnostic outcomes (r = 0.482, p < 0.01; β = 0.421, p < 0.001). Nearly half of respondents acknowledged that AI enhanced diagnostic accuracy and operational efficiency, indicating a moderate positive impact. These findings resonate with broader research showing that AI can outperform or complement human clinicians in tasks such as imaging interpretation, anomaly detection, and pattern recognition ([8]; [4]). For example, AI models have achieved diagnostic performance comparable to radiologists in detecting pulmonary nodules and diabetic retinopathy, often with faster processing times and reduced variability ([11]; [27]). Such evidence supports the study’s finding that AI enhances speed and accuracy where it is actively integrated. However, the moderate impact reported in this study contrasts with some higherresource settings where AI adoption is more mature. In Ghana, the limited scope of implementation, challenges in data quality, and lack of seamless integration with clinical systems likely contribute to the tempered perception of AI benefits ([20]; [3]). Moreover, while AI algorithms can assist in processing large volumes of data, clinicians have expressed concerns about generalizability and clinical relevance, particularly when systems are trained on nonrepresentative data ([15]; [33]). The ANOVA results showing differences in AI impact perceptions across professional roles highlight how clinical exposure and training influence acceptance and perceived utility. Existing research suggests that professionals who interact more frequently with digital tools, such as radiologists and senior physicians, are more likely to recognize AI’s benefits than those with limited exposure ([14]; [32]). This underscores the need for tailored capacity building and training programs to broaden understanding and trust across all healthcare roles.

Challenges of AI Adoption in Ghanaian Hospitals

Respondents reported that inadequate infrastructure, limited training, high implementation costs, professional resistance, and insufficient system integration remain the most significant barriers. These findings are consistent with prior research on digital health adoption in LMICs. Inadequate infrastructure, particularly the lack of robust computing hardware, network connectivity, and interoperable electronic health records, has been widely documented as a key obstacle to AI implementation ([12]; [22]). Without foundational digital systems, advanced AI tools cannot operate effectively, which explains why respondents perceived infrastructure as the most pressing challenge. Limited technical expertise and training emerged as a major constraint, with many staff indicating they lack the skills to operate or interpret AI outputs. This echoes findings from similar research in Ghana and other African countries, where training gaps and low digital literacy hinder the effective use of health technologies ([2]; [21]). High implementation and maintenance costs were also cited, reflecting the financial pressures faced by public health facilities where competing priorities often limit investment in new technologies ([6]; [5]). Professional resistance, observed among a considerable proportion of respondents, aligns with global literature that identifies clinician scepticism and fear of job displacement as barriers to AI adoption ([19]; [29]). Accepting AI as an assistive rather than a replacement tool requires cultural and organizational shifts within clinical teams ([10]; [28]). Insufficient integration of AI systems with existing workflows was another challenge, reflecting the difficulty of embedding new tools into routine practice without disrupting service delivery. This constraint is consistent with findings that emphasize the importance of system interoperability, workflow redesign, and usercentered implementation strategies to realize AI’s benefits ([16]; Reddy et al., 2020). The significant differences in AI adoption between tertiary and regional hospitals, as shown in the chi-square analysis, further underscore resource disparities that affect technology uptake. Tertiary facilities with stronger research links and external funding are more likely to pilot and sustain AI initiatives ([22]; [2]).

Conclusion

This study examined the role of Artificial Intelligence (AI) in enhancing diagnostic decision-making in selected health facilities in Ghana, specifically Korle Bu Teaching Hospital and Greater Accra Regional Hospital. The findings demonstrate that AI adoption in Ghanaian hospitals is moderate, concentrated mainly in radiology and laboratory diagnostics, while integration into clinical decision support and patient monitoring systems remains limited. The study revealed that AI utilization is significantly associated with improvements in diagnostic accuracy, speed, and workflow efficiency, confirming its potential to enhance clinical outcomes. Regression and correlation analyses showed that higher AI adoption predicts better diagnostic performance, while inferential tests indicated that tertiary hospitals and clinicians with greater exposure perceive more benefits. Despite these positive effects, the study identified significant barriers to AI adoption, including inadequate infrastructure, limited technical expertise, high implementation costs, professional resistance, and poor integration with existing workflows. These challenges are more pronounced in regional hospitals, highlighting disparities in resource availability and access to technology. The study concludes that AI has strong potential to transform diagnostic practices in Ghana, but realizing its full benefits requires strategic investments in technology, workforce training, system integration, and supportive policies. Addressing these factors will ensure equitable access and maximize the impact of AI on patient care and diagnostic efficiency across the Ghanaian healthcare system.