A Comparative Study of Melanoma Images Using CNN And Resnet 50

Abstract

Melanoma is a specific type of skin cancer that can be lethal if not diagnosed and treated early. This paper presents a deep-learning approach for the automatic identification of melanoma on dermoscopic images from the ISIC Archive dataset and non-dermoscopic images from the MED-NODE dataset. The method involves the development of Convolutional Neural Network (CNN) and ResNet50 models, along with various pre-processing techniques. The CNN and ResNet50 models detect melanoma from dermoscopic images with 98.07% and 99.83% accuracy respectively, using hair removal and augmentation techniques. For non-dermoscopic images, the CNN and ResNet50 models achieve an accuracy of 97.06% and 100% respectively, using the hair removal technique. Furthermore, combining age and gender as additional factors in identifying melanoma in dermoscopic images, leads to an accuracy of 96.40% using CNN. The results of this research suggest that the developed models when combined with various pre-processing techniques and the integration of age and gender as additional factors, can be an efficient tool in the early detection of melanoma.

Tool wear monitoring based on scSE-ResNet-50-TSCNN model integrating machine vision and force signals

“…Comparison of different modelsTo assess the performance and effectiveness of the proposed scSE-ResNet-50-TSCNN model in terms of accuracy, this paper conducted a comprehensive comparative analysis involving several other models. The models subjected to comparison include the following: a single force signal branch network without attention mechanism optimization, Force-ResNet-50[27]; a single image branch network without attention mechanism optimization, Image-ResNet-50[28]; a two-stream network without attention mechanism optimization, ResNet-50-TSCNN[29,30]; a single force signal branch network optimized with attention mechanism, Force-scSE-ResNet-50; and a single image branch network optimized with attention mechanism, Image-scSE-ResNet-50. All…”