MMR-CNN-soft-NMS: An efficient wound segmentation algorithm for diagnosis of peripheral artery disease

Peripheral artery disease (PAD) is a disease that requires a lot of medical care yet is often disregarded by people. Till now, wound segmentation has been investigated by various deep learning techniques which failed to detect subtle patterns and achieve low segmentation accuracy. To support the medical professionals, this project primarily aims to construct a clinical decision support system (CDSS). Therefore, around 3329 clinical wound images are originated from a range of subjects, including those who had suffered generalized trauma as well as PAD are analyzed in this research. For the diagnosis of PAD, this study introduces a segmentation method known as modified mask region-based convolutional neural network (MMR-CNN) which enhances the accuracy particularly in severe occlusion. The feature extraction performance is being enhanced by using Res2Net in combination with a multi-scale backbone. Later, the mask prediction stage produces segmentation masks for each region proposal; to further refine these masks, soft non-maximum suppression (soft-NMS) with an attenuation function is used which enhances the detection efficacy in cases with large overlap rates and the capacity. Overall, both Res2Net and soft-NMS significantly improve segmentation and feature extraction performance. The proposed MMR-CNN-soft-NMS offers a precise and effective way to identify and categorize wounds, especially in patients with PAD. When compared with other CNN models, the proposed MMR-CNN-soft-NMS model had better results including recall (0.91), accuracy (0.90), precision (0.90), and F1 score (0.89), respectively. From the analysis, the proposed MMR-CNN-soft-NMS accomplishes an improved diagnostic accuracy and supports in several medical imaging applications.