Artificial intelligence (AI) in the form of machine learning has successfully been applied to image-based diagnostics within several medical fields [1]. Deep learning is a machine learning method that utilizes artificial neural networks that mimic the neurons of the brain and enables computers to learn and represent complex patterns in data. Convolutional neural networks (CNNs) and vision transformers are currently the most commonly used artificial neural networks for image classification and interpretation [2]. A strength of CNNs is their ability to extract features from images, such as edges or textures, which they can then use to identify complex patterns like nuclei, parasite eggs, or other structures [3]. In contrast, vision transformers split images into patches and can identify dependencies and relationships between these patches [2]. Combinations of CNNs, transformers, and other AI models have been used in attempts to leverage the strengths of different architectures [3].

Implementing AI-based diagnostics into the workflow has the potential to automate processes, increase productivity in laboratories, and improve diagnostic accuracy [4]. Multiple AI-based diagnostic systems have been approved by the US Food and Drug Administration (FDA) or corresponding authorities in Europe (notified bodies for CE marking), for example, for cervical cancer screening and prostate cancer diagnostics [4-6]. Most diagnostic systems in use utilize high-end digital imaging instruments, so called whole-slide scanners, and require access to advanced laboratory infrastructure and may, therefore, not be optimal for use in primary health care (PHC) laboratories [4,5]. However, the development of cheaper, portable digital microscope scanners has enabled research on the use of AI-supported diagnostic systems suitable for PHC laboratories [7,8].

The World Health Organization (WHO) has emphasized the importance of providing diagnostics near the patient in PHC settings, to enhance the accuracy and timeliness of diagnoses, improve clinical decision-making, and reduce the risk of diagnostic errors [9,10]. A PHC laboratory, also known as a tier 1 laboratory, can be defined as a laboratory primarily serving outpatients by providing point-of-care tests, performing slide microscopy for simple preparations, and preparing fine needle aspirations and other simple tissue specimens that are later dispatched to a tier 2 laboratory for analysis. The tier 1 laboratories work with a small budget compared to more advanced laboratories and are generally managed by a laboratory technician, supervised by a pathologist from a distance [10].

The implementation of AI-supported diagnostic systems could be particularly advantageous at PHC laboratories. To begin with, since PHC laboratories do not have access to expertise in the form of a pathologist, the application of AI could enable further analyses being performed on site, consequently increasing the availability of diagnostics [10,11]. In addition, a systematic review showed the implementation of AI-supported diagnostics for microscopy increased the effectivity of laboratory personnel [4]. Henceforth, implementation of AI could potentially lower the costs of diagnostics. Furthermore, moving diagnostics from advanced laboratories to PHC laboratories has the potential to enable faster diagnostics.

Studies have investigated the use of AI-supported microscopy in PHC settings in diagnostics for different diseases, such as oral and cervical cancers [7,12]. Furthermore, studies have targeted various parasitic infections, for example, schistosomiasis and infections caused by soil-transmitted helminths [13,14]. Although the targeted diseases differed in these studies, it appears that the researchers faced similar challenges because of the commonalities in the methodologies applied. Challenges observed when comparing these studies include the following: first, the sample preparation method needs to be simple enough to be easily performed in PHC laboratories, while maintaining sufficient sample quality, to enable AI-based analysis of the digitized sample. For example, variance in sample thickness and preparation artifacts can result in poor-quality, whole-slide images, which makes analysis with AI models challenging [8,13]. Second, the sample digitization instrument needs to be cost-efficient and easy to use in a PHC setting (with potentially unstable internet and electricity access, as well as other demanding environmental conditions) [8]. Third, the AI models need to be reliable and feasible to implement in the diagnostic workflow in a way that provides robust and accurate diagnostics. This puts a demand on acquiring training data of sufficient size and variance, as well as using appropriate AI models.

A preliminary search of the databases PubMed and Cochrane was performed to investigate whether any scoping or systematic review had been performed investigating the use of AI-supported digital microscopy in PHC laboratories. A few similar reviews were found. A systematic review of AI diagnostics for oral cancer [15] has some overlap with our proposed review, but because it investigates a single disease, it does not provide an overview of the development of AI-supported digital microscopy in PHC laboratories. A systematic review evaluating the application of AI to whole-slide images of tissue samples stained with hematoxylin and eosin (H&E) was also identified [16]. This article presents the current state of knowledge on AI implementation in pathology in high-end laboratories, highlighting different approaches regarding datasets, preprocessing of images, and different approaches to image analysis. However, to our knowledge, no scoping review has been performed to compile the present knowledge and evidence on AI-supported digital microscopy diagnostics in PHC settings.

A scoping review performed on this subject is of particular value, since multiple challenges exist that needs to be mapped and addressed to successfully implement a diagnostic system with AI-supported digital microscopy in PHC laboratories. Such a review can also provide knowledge regarding what approaches have been applied so far and guide future research toward a potential implementation of AI-supported digital microscopy in PHC laboratories.

This scoping review aims to systematically review published studies related to AI-supported digital microscopy in PHC laboratories. The scoping review will specifically address the following questions: (1) In which diseases has AI-based microscopy been applied for diagnostics within PHC laboratories? (2) What methods have been used in acquiring microscopy images to train and analyze AI models for diagnostics? (3) What AI models and training approaches have been applied? and (4) How has the AI-supported diagnostic system performed compared to expert microscopists with regard to diagnostic accuracy?

The research question was as follows: What studies have been published on implementing AI-supported digital microscopy in PHC laboratories? What methods have been used, what issues have been faced, and what results have been achieved?

The proposed scoping review will be conducted in accordance with the JBI methodology for scoping reviews [17]. A PRISMA-ScR (Preferred Reporting Items for Systematic reviews and Meta-Analyses extension for Scoping Reviews) checklist can be found in Multimedia Appendix 1 [18]. A protocol has been published in the Open Science Framework (OSF) [19]. The inclusion and exclusion criteria are presented in Table 1.

All types of diagnostic test accuracy studies will be included. Because diagnostic test accuracy studies can be both retrospective and prospective, studies using either approach will be included. Additionally, studies using both paired and random designs for reference standards will be included [20]. The included studies must be published in English.

The search strategy was designed to identify peer-reviewed published articles. An initial limited search of PubMed and Cochrane was undertaken to identify articles on the topic. Search blocks were created for the final search based on terms used in the identified articles. The search blocks were developed to find articles containing the 2 concepts, microscopy and AI, as well as the context specification of being in a PHC setting, with 1 block created for each. A detailed description of the search strategy is presented in Multimedia Appendix 2. The reference lists of all included articles will be reviewed for additional studies and duplicates. In addition, all the articles citing the included articles will be reviewed. The databases to be searched are PubMed, Web of Science, Embase, and IEEE.

Following the search, all identified articles will be compiled in a reference management software system (Zotero, version 6.0.20; January 13, 2023; Corporation for Digital Scholarship), and duplicates will be removed. Following a pilot test, titles and abstracts will be screened by 2 independent reviewers for assessment against the inclusion criteria of the review. After the exclusion based on abstracts, the full text of the remaining studies will be assessed in detail against the inclusion criteria by 2 independent reviewers. The reasons for exclusion during the full-text screening will be recorded and reported. Any disagreements that arise between the reviewers at each stage of the selection process will be resolved through discussion between the reviewers and an additional researcher. The results of the search and the study inclusion process will be reported in full in the final scoping review and presented in a PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) flow diagram (Figure 1) [21].

Data will be extracted from studies included in the scoping review by 2 independent reviewers using a data extraction tool developed by the reviewers, which can be found in Multimedia Appendix 3. Any disagreements that arise will be solved through discussion between the reviewers and an additional researcher. If any uncertainties in the data arise, the corresponding author of the manuscript will be contacted. The findings will be presented narratively and in a table format based on the extraction tool developed beforehand by the researchers. The table will have eight columns: (1) the study name, year of publication, and authors; (2) what disease the study investigated; (3) dataset preparation; (4) training procedure and dataset; (5) AI model architecture; (6) QUADAS-2 (Quality Assessment of Diagnostic Accuracy Studies-2) risk of bias; (7) validation set, results, and reference standard; and (8) additional comments. The column of dataset preparation will contain sample collection and preparation as well as digitization procedure. The narrative presentation of the results will aim to provide an overview of the results.

To investigate the bias of the included studies, the QUADAS-2 tool will be applied. This tool was developed to assess the risk of bias for primary diagnostic accuracy studies in 4 areas: patient selection, index test, reference standard, and flow and timing [22]. The results will be presented in a table in the Results section, and the specific form for each article can be found in Multimedia Appendix 4.