JMIR Publications

ABSTRACT

Background:

Digital microscopy combined with artificial intelligence (AI) is increasingly being implemented in health care, predominantly in advanced laboratory settings. However, AI-supported digital microscopy could be especially advantageous in primary health care settings, since such methods could improve access to diagnostics via automation and a decreased need for experts on-site. To our knowledge, no scoping or systematic review has previously examined the use of AI-supported digital microscopy in primary health care laboratories, and a scoping review could guide future research by providing insights into the challenges of implementing these novel methods.

Objective:

This scoping review aimed to map published peer-reviewed studies on AI-supported digital microscopy in primary health care laboratories to generate an overview of the subject.

Methods:

A systematic search of the databases PubMed, Web of Science, Embase, and IEEE was conducted on October 2, 2024. The inclusion criteria in the scoping review were based on three concepts: using digital microscopy, AI, and comparison of the results to a standard diagnostic system; and one context, being performed in primary health care laboratories. Additional inclusion criteria were peer reviewed diagnostic accuracy studies in English performed on humans and achieving a sample level diagnosis. The study selection and data extraction were performed by two independent researchers, and cases of disagreement were solved through discussion with a third researcher. The methodology is in accordance with the JBI methodology for scoping reviews.

Results:

A total of 3,403 articles were screened during the article identification process, of which 22 (0.6%) were included in the scoping review. The samples analyzed were as follows: blood (n=12) for blood cell and malaria detection; urine (n=4) for urinalysis and parasite detection; cytology of atypical oral (n=1) and cervical cells (n=2); stool (n=2) for parasite detection; and sputum (n=1) for ferning-patterns indicating inflammation. Both conventional (n=15) and specifically developed methods (n=7) were used in sample preparation. The AI models used were based on both single (n=11) and multiple AI-algorithms (n=11) and all studies except one used convolutional neural networks. The AI-supported digital microscopy achieved comparable diagnostic accuracy to the reference standard for complete blood counts, malaria detection, identification of stool and genitourinary parasites, screening for oral and cervical cellular atypia, detection of pulmonary inflammation, and urinalysis. The AI-supported digital microscopy had higher sensitivity than manual microscopy in six out of seven (85.7%) studies that used a reference standard that allowed for this comparison.

Conclusions:

AI-supported digital microscopy achieved comparable diagnostic accuracy to the reference standard for diagnosing multiple targets in primary health care laboratories and may be particularly advantageous for improving diagnostic sensitivity. However, many shared challenges, ranging from sample preparation to workflow integration, need to be addressed to enable real world implementation. Clinical Trial: JMIR Res Protoc 2024;13:e58149 doi:10.2196/58149