Digital and Artificial Intelligence-based Pathology: Not for Every Laboratory - A Mini-review on the Benefits and Pitfalls of Its Implementation.

Abstract

Background and objectives: With the increasing use of artificial intelligence (AI) in diagnostics, AI algorithms have shown great potential in aiding diagnostics. As more of these algorithms are developed, there is overwhelming enthusiasm for implementing digital and artificial intelligence-based pathology (DAIP), but doubts and pitfalls are also emerging. However, few original or review articles address the limitations and practical aspects of implementing DAIP. In this review, we briefly examine the evidence related to the benefits and pitfalls of DAIP implementation and argue that DAIP is not suitable for every clinical laboratory.

Methods: We searched the PubMed database using the following keywords: "digital pathology," "digital AI pathology," and "AI pathology.". Additionally, we incorporated personal experiences and manually searched related papers.

Results: Ninety-two publications were found, of which 24 met the inclusion criteria. Many advantages of DAIP were discussed, including improved diagnostic accuracy and equity. However, several limitations of implementing DAIP exist, such as financial constraints, technical challenges, and legal/ethical concerns.

Conclusions: We found a generally favorable but cautious outlook for the implementation of DAIP in the pathology workflow. Many studies have reported promising outcomes in using AI for diagnosis and analysis; however, there are also several noteworthy limitations in implementing DAIP. Therefore, a balance between the benefits and pitfalls of DAIP must be thoroughly articulated and examined in light of the institution's needs and goals before making the decision to implement DAIP. Approaches for mitigating machine learning biases were also proposed, and the adaptation and growth of the pathology profession were discussed in light of DAIP development and advances.