Key points are not available for this paper at this time.
Background: Recently, the rapid advancement in generative artificial intelligence (AI) technology, such as ChatGPT-4, has sparked discussions, particularly in image recognition. Accurate results are critical for hematological diagnosis, particularly for blood morphology identification. Despite advanced hematology analyzers, reliance on professional hematopathologists for manual identification remains in cases of abnormal or rare conditions, a process prone to human subjectivity and potential errors. Consequently, this study aimed to investigate the potential of ChatGPT-4 to assist with blood morphology identification. Methods: We conducted a retrospective study using blood images obtained from the American Society of Hematology (ASH). These images comprised a range of normal and abnormal morphologies. Each sample was analyzed by expert technicians (control group) and classified using ChatGPT-4 (test group). Results: Preliminary results showed that ChatGPT-4 could identify normal blood cells with an accuracy of 88%, exceeding the accuracy of identifying abnormal blood cells at a rate of 54%. Regarding identifying abnormal cells, the accuracy of ChatGPT-4 was slightly higher than that of the manual method, which was 49.5%. Conclusion: This study shows that although generative AI shows the potential for blood type identification, it has not yet reached the point where it can replace the professional judgment of medical staff. The results showed that ChatGPT-4 is excellent for identifying red blood cell morphology, particularly inclusion bodies. It can be used as an auxiliary tool for clinical diagnosis. However, the overall recognition accuracy must be further improved. Our study produced innovative results in this field, establishing a foundation for future studies and highlighting the potential of generative AI in aiding blood morphology recognition. Future research should focus on enhancing the effectiveness of AI to improve overall standards of medical care.
Yang et al. (Tue,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: