ACDSSNet: Atrous Convolution-Based Deep Semantic Segmentation Network for Efficient Detection of Sickle Cell Anemia.

Area of Science:

• Medical Image Processing
• Deep Learning
• Computational Biology

Background:

• Semantic segmentation is crucial for precise anomaly localization in medical imaging.
• Challenges include noise, variations in cell shape/size, and viewpoint, complicating accurate segmentation.

Purpose of the Study:

• To propose novel Atrous Convolution-based Deep Semantic Segmentation Networks (ACDSSNet-I, ACDSSNet-II) for enhanced Sickle Cell Anemia (SCA) detection.
• To improve feature extraction, segmentation robustness, and boundary refinement in medical images.

Main Methods:

• Developed two novel Atrous Convolution-based Deep Semantic Segmentation Networks (ACDSSNet-I, ACDSSNet-II).
• Employed Atrous convolution-based dense prediction and Atrous spatial pyramid pooling for improved feature extraction and robust segmentation.
• Integrated efficient decoder modules and modified DeepLabV3+ architectures (MDA) using MobileNetV2 or ResNet50.
• Hybridized MDA-1 and MDA-2, optimizing with MobileNetV2, ADAM, and SGDM optimizers, and utilizing image saturation information.

Main Results:

• The proposed models achieved superior semantic segmentation performance.
• Achieved 98.21% accuracy, 99.00% specificity, and a 0.9547 Dice Similarity Coefficient (DSC).
• The hybridization strategy and optimal threshold selection further enhanced segmentation accuracy.

Conclusions:

• The novel ACDSSNet models offer significant improvements for SCA detection via semantic segmentation.
• The approach effectively mitigates challenges associated with noise and variations in medical images.
• The proposed methods demonstrate state-of-the-art performance in medical image analysis for disease detection.