Automated landmarking for insects morphometric analysis using deep neural networks

Abstract

Landmarks are one of the important concepts in morphometry analysis. They are anatomical points that can be located consistently (e.g., corner of the eyes) and used to establish correspondence or divergence among morphologies of biological or non-biological specimens. Currently, the landmarks are mostly positioned manually by entomologists on numerical images. In this work, we propose a method to automatically predict the landmarks on entomological images based on Deep Learning methods, more specifically by using Convolutional Neural Network (CNN). We propose a CNN architecture, EB-Net, which is built in a modular way the concept of "Elementary Blocks", each made up of usual layer types of CNN. After using a custom data augmentation procedure, the network has been trained and tested on a data set of different anatomical part of carabids (pronotum, head and elytra). In this numerical experiment, we have generated two strategies to evaluate the network and to improve the obtained results: training from scratch or applying a fine-tuning step. The predicted landmark coordinates have been compared to the coordinates of the manual landmarks provided by the biologists. The statistical analysis of the distances between predicted and manual coordinates has shown that our predictions can replace efficiently manual landmarking and allows to propose automatization of such operation.

Thirty years of geometric morphometrics: Achievements, challenges, and the ongoing quest for biological meaningfulness

“…(c) CCA maximizes the correlation between the latent variables, and the coefficients are proportional to multiple regression coefficients for both blocks (e.g., Fruciano, 2016;Menéndez, 2017;Waltenberger et al, 2021). In the last few years, numerous algorithms and software implementations for automated landmarking have been published (Aneja et al, 2015;Bannister et al, 2020;Bromiley et al, 2014;Devine et al, 2020;Galvánek et al, 2015;Le et al, 2020;Li et al, 2017;Percival et al, 2019;Porto et al, 2021;Porto & Voje, 2020;Vandaele et al, 2018).…”

“…It is also prone to error (e.g., Fruciano, 2016; Menéndez, 2017; Waltenberger et al, 2021). In the last few years, numerous algorithms and software implementations for automated landmarking have been published (Aneja et al, 2015; Bannister et al, 2020; Bromiley et al, 2014; Devine et al, 2020; Galvánek et al, 2015; Le et al, 2020; Li et al, 2017; Percival et al, 2019; Porto et al, 2021; Porto & Voje, 2020; Vandaele et al, 2018). Personally, we have no experience with these approaches, but the publications and what we have heard from colleagues appear promising.…”

Phenotypic systems biology for organisms: Concepts, methods and case studies

“…5 c). It notes that which points on morphological shapes are representative are, at least for now, due to biological knowledge, though quantitative methods (e.g., deep learning techniques [ 57 , 58 ]) would be more suitable to identify landmarks without prior pieces of knowledge.…”

A Lightweight Keypoint Matching Framework for Morphometric Landmark Detection

“…Few models are proposed for morphometric landmark detection in insect wing images because of the limited size of the dataset. Le et al 14 used three convolutional models to predict the morphometric landmarks on 260 beetle images; some augmentation techniques are used to increase the size of the dataset but overfitting still appears in two models.…”

“…5000) around each landmark then train an Extremely Randomized Tree classifier to decide if a candidate point is a landmark or not. Second, the proposed method needs only a few sample images (3)(4)(5)(6)(7)(8)(9)(10)(11)(12)(13)(14)(15) to learn while 9 and other methods require hundreds of images. Third, the proposed method requires fewer candidate points in the prediction stage.…”