The uncertainty in predicting what the future holds for blood cell counting can be best visualized by thinking how difficult it would have been to predict the present situation 50 years ago! However, there will be the dichotomy of building better cell counters for improved diagnostics versus smaller, more portable, or even wearable or implantable in-vivo devices. Such implantable or wearable devices can predict the disease status by monitoring hematological parameters in real time.
Artificial intelligence (AI) is the use of computer algorithms to perform tasks or solve problems that were traditionally solved using the adaptable and flexible human intelligence. Indeed, as of 2020, radio-diagnostics is the field where AI has made the most progress to date; 90 percent of the practicing radiologists anticipate that AI will be incorporated in their future practices. AI has also been employed in the analysis of hematopathology data for better informed diagnosis, prognosis, and treatment planning. The foundation knowledge related to benign and malignant hematology is also evolving with the application of AI in hematology.
Even the most skilled hematology specialist can overlook the patterns, deviations, and relations among the vast number of blood parameter measures by modern analyzers. In contrast, the AI algorithm can easily analyze hundreds of attributes (parameters) simultaneously with correlation, and recognize a pattern or fingerprint of certain hematological abnormality. The AI algorithms of modern hematology analyzers are trained on sufficiently large datasets of clinical cases that include laboratory blood tests, performed and confirmed by hematology specialist using various confirmatory tests. The machine learning (ML) algorithms can uncover the disease-related patterns in a blood test that may be beyond medical knowledge, resulting in higher diagnostic accuracy as compared to traditional quantitative interpretation, based only over reference ranges. Also, it can lead to a fundamental change in differential diagnosis and result in the modifications in the presently accepted guidelines.
Recently, Tabe and Kimura et. al. (Japan) have published their findings (Nature Scientific Reports Sept’19) where they have successfully developed and tested a deep learning-based algorithm – convolutional neural networks (CNN). This system is applied for differentiating myelodysplastic syndrome (MDS) from aplastic anemia (AA) with 96.2 percent of sensitivity and 100 percent specificity. This is the first such system for MDS, using peripheral blood smears; it can further be developed for automated diagnosis of various hematological abnormalities. This system recognizes over 100 patterns in cell size and cytoplasmic morphological features for detection and classification of myeloid malignant cells including MDS. Further training of this AI algorithm can improve the accuracy and reproducibility, which may surpass the human eye.
An understanding of AI’s basics will need to become a part of physician’s statistical literacy in the very near future. AI and ML algorithms and their application in hematology show a great promise in medical laboratory diagnoses. It will be of high value not only for laboratory physicians and patients, but will also have widespread beneficial impacts on healthcare costs!