Intelligent Multiclass Skin Cancer Detection Using Convolution Neural Networks

Recent advances of artificial intelligence in melanoma clinical practice

“…Then the GoogLeNet team carried out further exploration and improvement, resulting in Inception-v2, v3 [44], and v4 [45]. Compared with Inception-v1, Inception-v2 and v3 can achieve higher accuracy in melanoma diagnosis due to the smaller kernel size (3*3) convolution [15]. Inception-v4 follows the structure of Inception-v2/ v3.…”

“…More importantly, the combination of Inception with ResNet network, the Inception-ResNet The table is sorted by melanoma identification accuracy. network, could have higher accuracy with fewer epochs [15].…”

“…Though a variety of AI models are applied to melanoma diagnosis, in genuine AI training, doctors and researchers frequently use AlexNet, Inception, ResNet, and DenseNet [58]. Among them, AlexNet has the greatest melanoma detection accuracy rate (97.7%), and Inception-v3 is closely behind (97.0%) [14,15]. In practice, however, AlexNet will consume more memory due to its 12 times larger parameters than Inception [45].…”

“…Since it has more Inception modules, v4 speeds up the operation rate. More importantly, the combination of Inception with ResNet network, the Inception-ResNet network, could have higher accuracy with fewer epochs [15].…”

“…Both the following two studies use the GoogLeNet architecture to distinguish between melanoma and nevi. The technique was proposed by Alabduljabbar and Alshamlan and achieved a high accuracy of 97.0% for the dataset ISBI2016 [15]. This study used the Inception-v3 module, which has been trained on dermoscopic images (Fig.…”