Analyzing and visualizing morphological features using machine learning techniques and non‐big data: A case study of macaque mandibles

Abstract

Objectives: Morphometrics has played essential roles in the comprehension of biological variation and the evolution of morphological phenotypes. This approach usually imposes strict requirements on data, such as rigid alignment of subjects, and the collection and manual preprocessing of data meeting these requirements are often time consuming. Artificial intelligence (AI) technology is developing and it potentially reduces this load, but they usually presuppose the availability of "big data" for successful learning, beyond the empirically plausible amount in biological studies. Here, we propose a deep learning-based analysis of three-dimensional data. Materials and Methods:We built a deep learning-based analysis of three-dimensional morphological data that does not require strict alignment or an implausible sample size. We benchmarked the proposed method by case studying sex classification of macaques, referring to computed tomography scans of their mandible. Results:The model learned from just 139 mandible specimens of Japanese macaques and successfully generalized the learned classification to previously unseen specimens of the same species and even other species of macaques. Moreover, we visualized those characteristic regions in the data that the model used during sex classification and showed that they were consistent with the criteria used by human experts.Discussion: Our analysis does not require rigidly aligned data, so can effectively use data collected in previous studies with different focus/aims. This proposed AI method can potentially help researchers to discover new morphological features of different species and other biological groups. Implementation of this proposed AI system will be available to other researchers for further investigation.