Pharmacognostic analysis and teachable machine automated identification of Uvae ursi folium, Vitis idaeae folium and Buxi folium

Context: Microscopic analysis is a method in Pharmacognosy utilized in identifying herbal substances. Machine learning and computer vision are contemporary approaches leveraged in the identification of plants. Bearberry leaf is an active ingredient of medicinal products used to treat mild infections of the lower urinary tract. In the present study, we examine the unique microscopic diagnostic characteristics of Uvae ursi folium and two of its adulterants - Vitis idaeae folium and Buxi folium and evaluate the effectiveness of teachable machine models in automated microscopic preparation identification. Aims: To examine Uvae ursi folium, Vitis idaeae folium and Buxi folium micromorphology and micrometry and evaluate the effectiveness of teachable machine models in their microscopic identification. Methods: Microscopic preparations were mounted in 4 g/mL chloral hydrate aqueous solution. Photographic materials were captured using a Zeiss Primostar iLED microscope and analyzed in Labscope 4.2.1. Models were trained on datasets of photographic materials, compared by their intrinsic parameters (accuracy per class, confusion matrix, accuracy per epoch, and loss per epoch), and tested on datasets of 10 unseen images. Results: Microscopic descriptions, illustrations, and photographic materials were created for all herbal substances. Vitis idaeae folium and Buxi folium covering and glandular trichomes were described. Teachable machine models successfully distinguish between and identify the herbal substances. The most accurate models correctly identify all tested Uvae ursi folium photographic materials. Conclusions: Vitis idaeae folium and Buxi folium diagnostic characteristics are described, and teachable machine model capabilities in automated microscopic photographic material identification are explored for the first time.