Image Processing Techniques in Localization of DFU

EasyChair Preprint 12582

Abstract

For the detection and identification of diabetic foot ulcers (DFUs), there has been a significant amount of study using computer methodologies and technology; nevertheless, there are few systematic comparisons of cutting-edge deep learning techniques. Frameworks for object detection were used to solve this issue. A thorough dataset made up of 2,000 photos for training and 2,000 images for testing was made available to participants by DFUC2020. By contrasting the deep learning-based algorithms proposed by the winning teams, including

Faster R-CNN, three variations of Faster R-CNN, and an ensemble technique, YOLOv3 and YOLOv5, EfficientDiet, and a novel Cascade Attention Network, this paper summarises the results. We give a thorough explanation of the model architecture, training parameter values, and additional processes like pre-processing, data augmentation, and post-processing for each deep learning technique

We offer a detailed assessment of each technique. The number of photos available for training had to be increased with data

augmentation for each approach, and false positives had to be eliminated using post-processing. Deformable Convolution, a Faster R-CNN variation, delivered the greatest results, with an F1-Score of 0.9617 and a mean average precision (map) of 0.9565, recall:0.9670,accuracy:0,9510 Lastly, we show that the F1-Score can be improved using the ensemble method based on several deep learning techniques, but not the map. Convolutional neural network (CNN), deep learning, mobile net model, minutiae, ulcers in diabetic feet.

Keyphrases: Convolutional Neural Network (CNN), Keywords: Diabetic Foot Ulcers, Minutiae, Mobile Net Model, deep learning

@booklet{EasyChair:12582,
  author    = {Siva Praveen Lattala and Likithchowdar Yarlagadda and Koppineedi Ganesh Kumar and Bollu Vinay Chowdary and Gaurav Varshney},
  title     = {Image Processing Techniques in Localization of DFU},
  howpublished = {EasyChair Preprint 12582},
  year      = {EasyChair, 2024}}