Biology subjects

Sadia, H.

Publications and source records attributed to Sadia, H..

4 recordsLinked to original sources

bioRxiv2026

Deep-learning based 3D segmentation of heterogeneous lizard claw tissue from CT data

The accurate segmentation of lizard claws is important as they are materially heterogeneous, comprising both bone and keratinous tissue. This study presents a deep learning framework for the automated segmentation of lizard claw tissues, specifically bone and keratin, from CT imaging data. A dataset comprising 14 lizard claws was used in this work, with annotations generated through a superpixel based labeling approach to provide ground truth reference segmentations. To evaluate the effect of spatial context on segmentation performance, both 2D and 2.5D CNN architectures using DeepLabV3 with ResNet-50, ResNet-101, and Inception-ResNet-v2 backbones were investigated, with predictions subsequently reconstructed into three-dimensional volumes for analysis. Performance was assessed using a leave one out cross validation (LOOCV) strategy and evaluated with 3D Dice Similarity Coefficient (DSC), Intersection over Union (IoU), Sensitivity (Recall), 95th Percentile Hausdorff Distance (HD95), and Relative Volume Error (RVE). Experimental results demonstrate that 2.5D CNN architectures consistently outperform their 2D counterparts across all evaluation metrics, highlighting the importance of incorporating inter-slice contextual information for volumetric tissue segmentation. From amongst the models, the 2.5D Inception-ResNet-v2 achieved the best overall performance, reaching a validation accuracy of 97.5% and producing segmentation results that closely align with ground-truth tissue structures. Our findings demonstrate the effectiveness of 2.5D deep learning approaches for the high accuracy segmentation of heterogeneous lizard claw tissues from CT data, whilst providing a robust framework for automated morphological analysis in comparative anatomical studies.

zoology↗

bioRxiv2026

AI-BioMech: Deep Learning Prediction of Mechanical Behavior in Aperiodic Biological Cellular Materials

We introduce AI-BioMech, a deep learning based framework that directly predicts the mechanical response of cellular structures from 2D images, eliminating the need for manual geometry definition and traditional finite element simulations. The framework is trained on synthetic datasets representing biological cellular structures and benchmarked against real experimental data. Finite element analysis (FEA) based labeling is used to generate pixel level annotations for semantic segmentation, enabling accurate identification of stress and strain distributions. By learning spatial and hierarchical patterns from these annotations, the model automatically extracts complex features to predict cellular material responses under compressive loading conditions. Transfer learning with fine tuning by using the DeepLabv3 architecture with ResNet50, ResNet101, and Inception ResNetV2 backbones enhances prediction accuracy and generalization from limited datasets. Model predictions are validated against experimental results and Digital Image Correlation (DIC) measurements, demonstrating strong agreement with physical observations. The results show that AI-BioMech achieves up to 99% prediction accuracy while significantly outperforming traditional methods in computational speed and scalability.

biophysics↗

bioRxiv2025

A Standardized Method for Insect Color Analyses using Open Source Software: AInsectID Version 1.1 Color Merge

The accurate representation of color is important in applications involving species identification. Environmental variations introduce inconsistencies in color perception, affecting the reliability of automated image processing algorithms. In previous work, we developed a hybrid algorithm, AInsectID Version 1.1 Color Merge, to overcome challenges posed by over-segmentation and under-segmentation in insect wing color clustering. We achieved this by using color differences between superpixels to measure homogeneity during the superpixels segmentation process. Nevertheless, our algorithm remains sensitive to environmental effects, affecting its performance and accuracy in color analyses. Here, we introduce a standard imaging method for insect species, as a pre-requisite to analysis in AInsectID Version 1.1 Color Merge. We systematically examine the effects of varying lighting conditions, angle of observation, and working distance in a controlled environment to assess their impact on the performance of the algorithm. We find that by meticulously controlling lighting, working distance, and lighting angle, we develop an evidence-based standard approach to imaging colors that is robust and repeatable. By following our standardized procedure, consistent color analyses are possible under varying environmental conditions. The method was tested using the Delta E2000 ({Delta}E) color difference metric with a threshold of 1, demonstrating that our standard approach maintains perceptual accuracy within the Just Noticeable Difference (JND) range, while improving the reliability of color analyses of insect wings in diverse environments. Finally, to validate the robustness of our standardization method, we evaluated the certainty of our results at different levels of confidence.

zoology↗

bioRxiv2024

AInsectID Version 1.1: an Insect Species Identification Software Based on the Transfer Learning of Deep Convolutional Neural Networks

AInsectID Version 1.11, is a GUI operable open-source insect species identification, color processing2 and image analysis software. The software has a current database of 150 insects and integrates Artificial Intelligence (AI) approaches to streamline the process of species identification, with a focus on addressing the prediction challenges posed by insect mimics. This paper presents the methods of algorithmic development, coupled to rigorous machine training used to enable high levels of validation accuracy. Our work integrates the transfer learning of prominent convolutional neural network (CNN) architectures, including VGG16, GoogLeNet, InceptionV3, MobileNetV2, ResNet50, and ResNet101. Here, we employ both fine tuning and hyperparameter optimization approaches to improve prediction performance. After extensive computational experimentation, ResNet101 is evidenced as being the most effective CNN model, achieving a validation accuracy of 99.65%. The dataset utilized for training AInsectID is sourced from the National Museum of Scotland (NMS), the Natural History Museum (NHM) London and open source insect species datasets from Zenodo (CERNs Data Center), ensuring a diverse and comprehensive collection of insect species.

zoology↗