Artificial intelligence (AI) and machine learning based applications are thought to impact the practice of healthcare by transforming diagnostic patient management approaches. However, domain knowledge, clinical and coding expertise are likely the biggest challenge and a substantial barrier in developing practical AI models. Most informatics and AI experts are not familiar with the nuances in medicine, and most doctors are not efficient coders. To address this barrier, a few "no-code" AI platforms are emerging. They enable medical professionals to create AI models without coding skills. This study examines Teachable Machine™, a no-code AI platform, to classify white blood cells into the five common WBC types. Training data from publicly available datasets were used, and model accuracy was improved by fine-tuning hyperparameters. Sensitivity, precision, and F1 score calculations evaluated model performance, and independent datasets were employed for testing. Several factors that influence the performance of the model were tested. The model achieved 97% accuracy in classifying white blood cells, with high sensitivity and precision. Independent validation supported its potential for further development. This is the first study that demonstrates the value of a free no-code algorithm based AI platforms use in hematopathology using authentic datasets for training. It opens an opportunity for the healthcare professionals to get hands-on experience with AI and to create practical AI models without coding expertise.
Keywords: Hematology; Machine Learning; No-Code platforms; Peripheral Blood White Cell Differentials; Teachable Machine.
Copyright © 2024, Hoseini et al., Applied Systems and Discoveries Journals.