Image Segmentation and Classification for Sickle Cell Disease using Deformable U-Net (1710.08149v3)

Abstract: Reliable cell segmentation and classification from biomedical images is a crucial step for both scientific research and clinical practice. A major challenge for more robust segmentation and classification methods is the large variations in the size, shape and viewpoint of the cells, combining with the low image quality caused by noise and artifacts. To address this issue, in this work we propose a learning-based, simultaneous cell segmentation and classification method based on the deep U-Net structure with deformable convolution layers. The U-Net architecture for deep learning has been shown to offer a precise localization for image semantic segmentation. Moreover, deformable convolution layer enables the free form deformation of the feature learning process, thus makes the whole network more robust to various cell morphologies and image settings. The proposed method is tested on microscopic red blood cell images from patients with sickle cell disease. The results show that U-Net with deformable convolution achieves the highest accuracy for segmentation and classification, comparing with original U-Net structure.

• The paper proposes a Deformable U-Net architecture for simultaneous red blood cell segmentation and classification in sickle cell disease, addressing morphology variations.
• Integrating deformable convolution allows the network to adapt to irregular cell shapes, significantly improving segmentation accuracy to 97.8% and classification performance.
• This automated method offers a robust solution for handling complex cell morphologies in SCD images, showing potential for clinical application and more precise diagnosis.

The paper "IMAGE SEGMENTATION AND CLASSIFICATION FOR SICKLE CELL DISEASE USING DEFORMABLE U-NET" proposes a novel deep learning approach for the simultaneous segmentation and classification of red blood cells (RBCs) in sickle cell disease (SCD) using a modified U-Net architecture integrated with deformable convolution layers. The primary motivation is to address the substantial variability in RBC morphology due to SCD, which poses challenges for accurate and automatic segmentation and classification using conventional image processing techniques.

Introduction and Motivation

Sickle cell disease significantly alters RBC morphology, resulting in cells with heterogeneous shapes that challenge traditional segmentation methods such as thresholding, region growing, and watershed transform. These methods struggle with noisy backgrounds and blurred boundaries typical of microscopic images, necessitating a robust approach capable of handling these variations.

Methodology

Deformable U-Net Architecture

The paper builds on the U-Net architecture, a convolutional neural network (CNN) widely acclaimed for its semantic segmentation capabilities, further enhanced with deformable convolution. The deformable convolution enables adaptive receptive fields, effectively allowing the network to manage geometric variations in cell shapes and sizes.

Key Components:

• Deformable Convolution: Modifies the standard convolution operation by introducing learned offsets to the sampling grid, enhancing the network's flexibility in feature representation.
• Architecture Design: The deformable U-Net consists of an encoder-decoder structure where the encoder path captures features through down-sampling layers, while the decoder path reconstructs the segmentation maps. The use of deformable convolutions throughout the architecture provides it with rich spatial-awareness capabilities.

Data Handling

The dataset comprises microscopic RBC images from SCD patients, sampled at high resolution and manually annotated to facilitate supervised training of the models. Due to GPU memory constraints, the images are divided and resized to smaller samples for efficient processing.

Results and Evaluation

The experimental results demonstrate that the deformable U-Net outperforms the baseline U-Net in both segmentation and classification tasks. The performance is rigorously evaluated using measures such as loss, false negative (FN) rate, false positive (FP) rate, and classification errors (Error I and Error II).

Performance Highlights:

• Improved Segmentation Accuracy: Deformable U-Net achieves better segmentation accuracy (97.8%) compared to the baseline U-Net (94.7%) by effectively reducing false positives.
• Superior Classification Performance: For cell classification, the improved handling of cell boundary consistency leads to higher classification accuracy with reduced Error II rates.

Discussion

The paper highlights the deformable kernel's effectiveness in managing object edges and spatial variance, an advantage not fully realized by conventional U-Net architectures. The introduction of flexible receptive fields via deformable kernels significantly improves the model's capacity to discriminate between background noise and actual cell structures.

Conclusion and Future Work

The deformable U-Net presents a robust and efficient solution for automatic RBC segmentation and classification in SCD, showing potential for clinical application. Future efforts will focus on expanding the dataset to encompass a broader range of cell types, thereby enhancing the model's capability for fine-grained classification and aiding more precise SCD diagnoses. The paper suggests that despite the increase in computational demand during training, the deployment efficiency remains practical for real-world applications.