ConvNeXt-FD: A Fractal-Based Deep Model for Robust Biomedical Image Segmentation

2026-05-22 · Source: cs.CV updates on arXiv.org · Field: Science & Research — Health & Medical Research, Mathematics & Computational Sciences · Depth: Expert, extended

Summary

ConvNeXt-FD is a novel deep learning architecture designed for robust biomedical image segmentation, employing a U-Net-like encoder-decoder framework with a ConvNeXt backbone. This model integrates a hybrid loss function, combining the Dice coefficient with a boundary-aware regularization term inspired by Fractal Dimension, which enhances sensitivity to object boundaries and shape fidelity. Rigorously evaluated across six diverse biomedical datasets—BUSI, DDTI, FluoCells, IDRiD, ISIC2018, and MoNuSeg—ConvNeXt-FD demonstrates competitive and often superior performance against existing methods. Key findings include significant performance boosts, over 16% in Dice score for BUSI and DDTI, when using ImageNet pre-trained weights. The optimal weighting for the boundary-aware loss (λ_FD) is task-dependent, ranging from 0.01 to 0.50, proving its adaptability to varied morphological challenges.

Key takeaway

For machine learning engineers developing biomedical image segmentation models, you should integrate modern CNN backbones like ConvNeXt within U-Net architectures. Prioritize ImageNet pre-training to significantly boost performance, especially for datasets with subtle boundaries. Furthermore, consider implementing a hybrid loss function with a boundary-aware regularization term, such as the fractal-inspired approach, and carefully tune its weighting (λ_FD) to achieve superior shape fidelity and competitive results across diverse medical imaging tasks.

Key insights

ConvNeXt-FD enhances biomedical image segmentation by combining ConvNeXt with fractal-inspired boundary regularization for superior shape fidelity.

Principles

• Transfer learning from ImageNet significantly improves medical image segmentation.
• Boundary-aware regularization improves segmentation accuracy, especially for irregular shapes.
• Optimal loss weighting is task-dependent for diverse morphological challenges.

Method

ConvNeXt-FD uses a U-Net-like encoder-decoder with a ConvNeXt backbone. It optimizes with a hybrid loss: Dice coefficient + a boundary-aware term derived from a differentiable Fractal Dimension map.

In practice

• Apply ImageNet pre-training for medical image segmentation tasks.
• Experiment with boundary-aware loss functions for precise delineations.
• Tune regularization hyperparameters (λ_FD) based on dataset characteristics.

Topics

• Biomedical Image Segmentation
• ConvNeXt
• Fractal Dimension
• Deep Learning Architectures
• Hybrid Loss Functions
• Transfer Learning

Best for: Computer Vision Engineer, AI Scientist, Machine Learning Engineer, Research Scientist

Related on AIssential

• ROBUST-WT: Robust Uncertainty-aware Segmentation Transform via Whitening and Training Enhancements — ROBUST-WT demonstrates that targeted training enhancements can substantially improve medical image segmentation robustness without altering core architecture.
• Fourier fractal dimension to predict the generalization of deep neural networks — Fourier fractal dimension of network weight variations predicts deep neural network generalization without hold-out data.
• Federated Medical Image Segmentation under Real-World Label Noise: A Benchmark Suite for Noisy Label Learning Method Selection — The new benchmark suite enables realistic evaluation and selection of Federated Noisy Label Learning methods for medical image segmentation.
• Learning a Sampling-Free Variational DNN Plugin from Tiny Training Sets to Refine OOD Segmentation With Uncertainty Estimation — VarDeepPCA refines OOD medical image segmentation using intrinsic geometric priors and sampling-free variational learning from tiny in-distribution datasets.
• Multi-layer feature aggregation network with residual module and attention mechanism for jaw cyst image segmentation — MFA-Net enhances medical image segmentation by integrating residual modules, channel-spatial attention, and multi-layer feature aggregation.
• Quantification of Uncertainty with Adversarial Models in Medical Image Segmentation — QUAM-SM uses adversarial search to identify fragile pixels, disentangling uncertainty for reliable medical image segmentation.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.CV updates on arXiv.org.