ForamDeepSlice: A High-Accuracy Deep Learning Framework for Foraminifera Species Classification from 2D Micro-CT Slices

ForamDeepSlice: A High-Accuracy Deep Learning Framework for Foraminifera Species Classification from 2D Micro-CT Slices A Preprint Abdelghafour Halimi Visualization Core Lab Thuwal, 23955, Saudi Arabia abdelghafour.halimi@kaust.edu.sa Ali Alibrahim Physical Sciences and Engineering Division Thuwal, 23955, Saudi Arabia ali.alibrahim@kaust.edu.sa Didier Barradas-Bautista Visualization Core Lab Thuwal, 23955, Saudi Arabia didier.barradas@kaust.edu.sa Ronell Sicat Visualization Core Lab Thuwal, 23955, Saudi Arabia ronell.sicat@kaust.edu.sa Abdulkader M. Afifi Physical Sciences and Engineering Division Thuwal, 23955, Saudi Arabia abdulkader.alafifi@kaust.edu.sa December 2, 2025 Abstract This study presents a comprehensive deep learning pipeline for the automated classification of 12 foraminifera species using 2D micro-CT slices derived from 3D scans. We curated a scientifically rigorous dataset comprising 97 micro-CT scanned specimens across 27 species, selecting 12 species with sufficient representation for robust machine learning. To ensure methodological integrity and prevent data leakage, we employed specimen-level data splitting, resulting in 109,617 high-quality 2D slices (44,103 for training, 14,046 for validation, and 51,468 for testing). We evaluated seven state-of-the-art 2D convolutional neural network (CNN) architectures using transfer learning. Our final ensemble model, combining ConvNeXt-Large and EfficientNetV2-Small, achieved a test accuracy of 95.64%, with a top-3 accuracy of 99.6% and an area under the ROC curve (AUC) of 0.998 across all species. To facilitate practical deployment, we developed an interactive advanced dashboard that supports real-time slice classification and 3D slice matching using advanced similarity metrics, including SSIM, NCC, and the Dice coefficient. This work establishes new benchmarks for AI-assisted micropaleontological identification and provides a fully reproducible framework for foraminifera classification research, bridging the gap between deep learning and applied geosciences. Keywords foraminifera classification, deep learning, transfer learning, 2D slices, 3D micro-CT data 1 Introduction Paleontology is undergoing a transformative data revolution, catalyzed by the widespread adoption of high-resolution, non-destructive imaging technologies such as micro-Computed Tomography (micro-CT) [Knutsen and Konovalov, 2024, Edie et al., 2023]. These imaging methods have become indispensable for exploring the internal morphology of fossil specimens, offering unprecedented access to delicate anatomical structures that would otherwise be compromised by traditional mechanical preparation techniques [Heřmanová et al., 2020, Cunningham et al., 2014]. However, this arXiv:2512.00912v1 [cs.CV] 30 Nov 2025 Foraminifera Classification using Deep Learning A Preprint technological leap has introduced a significant bottleneck in post-processing: the manual segmentation of fossils from surrounding rock matrices. As imaging resolutions increase, the volumetric datasets grow exponentially, making segmentation—especially for low-contrast specimens like calcareous fossils in carbonate-rich matrices—the most labor-intensive and error-prone step in the workflow [Edie et al., 2023]. To address this challenge, the field has increasingly turned to deep learning, a subset of machine learning that excels in automated image analysis. Early applications demonstrated that convolutional neural networks (CNNs), particularly U-Net architectures, could segment fossils with accuracies rivaling manual annotation, but in a fraction of the time [Knutsen and Konovalov, 2024, Edie et al., 2023, Itaki et al., 2020a]. These breakthroughs have laid the foundation for a broader integration of artificial intelligence (AI) into paleontological research, extending beyond segmentation to include fossil classification and reconstruction, as summarized in Table 1. One of the most promising directions in this domain is the classification of microfossils using transfer learning. Several studies have shown that pretrained CNN architectures—such as VGG16, ResNet50, MobileNetV2, InceptionV3, and Xception—can be fine-tuned to classify microfossils from low-resolution optical images with remarkable accuracy. For instance, classification of the genus Globotruncanita achieved up to 96.97% accuracy and an AUC of 0.978 [Ozer et al., 2024], while broader genus-level classification reached 99.78% accuracy [Ozer et al., 2023a]. These approaches are particularly valuable for laboratories with limited computational resources, offering high performance with minimal infrastructure. Beyond transfer learning, researchers have explored custom architectures built from scratch, combining CNNs with recurrent neural networks (LSTM, BiLSTM) to classify species within the genus Globotruncana, achieving up to 97.35% accuracy and an AUC of 0.968 [Ozer et al., 2023b]. Classical machine learning models such as SVM, LDA, and kNN have also demonstrat

…(Full text truncated)…

This content is AI-processed based on ArXiv data.