ZenToks Publication

Chapter 3

Morphometric and Machine Learning-Based Classification of Vertebrate Blood Using Hemin Crystal Geometry

Abstract/Preface

Accurate taxonomic identification of vertebrate blood is essential in forensic science, wildlife crime investigation, and comparative biology. While molecular methods such as DNA barcoding provide high precision, they are often constrained by cost, infrastructure requirements, and processing time. This study evaluates whether quantitative morphometric analysis of hemin (Teichmann) crystals, a classical blood confirmation test, can serve as a rapid, low-cost tool for class-level taxonomic discrimination. Hemin crystals were prepared from blood samples representing five vertebrate classes (Mammalia, Aves, Reptilia, Amphibia, Pisces) under standardised conditions. Crystal length, breadth, and aspect ratio were measured using calibrated light microscopy. Classification models, including Linear Discriminant Analysis (LDA) and Random Forest (RF), were employed to predict taxonomic class based on morphometric variables. LDA achieved complete separation of taxonomic classes in train-test evaluation, with the first two discriminant axes accounting for 93.71% of between-class variance. Ten-fold cross-validation yielded a mean classification accuracy of approximately 90%. Random Forest classification independently corroborated these findings, achieving 100% test accuracy and identifying aspect ratio as the most influential predictor. These results demonstrate that hemin crystal geometry encodes systematic, taxonomically informative variation at the class level, likely reflecting underlying differences in haemoglobin structure and crystallisation behaviour across vertebrates. The integrated morphometric–machine learning framework provides a robust, reproducible, and cost-effective complementary approach for rapid blood screening in forensic and ecological contexts, particularly where molecular techniques are impractical.