Comparison of Inceptionv3, DenseNet201 and ResNet50 Convolutional Neural Networks for Skin Lesion Classification

Abstract

Skin cancer is one of the most common oncological diseases, accounting for approximately 5% of all cancer diagnoses, with a mortality rate reaching 2% of all cases [1]. Diagnosis typically involves visual inspection, dermatoscopy, histology, and machine learning methods. A meta-analysis of 104 studies showed that dermatoscopy increases diagnostic sensitivity by 34% at a fixed specificity of 80% [2]. The use of neural networks trained on dermatoscopic images can further improve diagnostic accuracy and reduce errors.
For this study, the HAM10000 dataset was used, containing 10,015 dermatoscopic images of seven types of skin lesions, with over 50% of diagnoses histologically confirmed [3]. Images were grouped into two categories: benign (8,061 images) and malignant (1,954 images). Preprocessing included normalization, augmentation (rotation, shift, scaling, flipping), and resizing to 200×150 pixels. For classification, DenseNet201, ResNet50, and InceptionV3 models with pretrained ImageNet weights were used. The top layers were replaced with GlobalAveragePooling, ReLU, BatchNormalization, Dropout, and softmax layers. The loss function was sparse_categorical_crossentropy.
DenseNet201 achieved the highest accuracy: 0.89 (training), 0.87 (validation). InceptionV3 reached 0.86 accuracy on both sets after 91 epochs. ResNet50 showed the lowest accuracy: 0.79 (training), 0.81 (validation) after 45 epochs. In summary, DenseNet201 demonstrated the highest efficiency for skin lesion classification. Convolutional neural networks trained on dermatoscopic images offer high potential in the diagnosis of skin neoplasms, increasing accuracy and sensitivity, and may serve as an effective adjunct to clinical expertise, especially for early detection of malignant skin lesions.

[1] Siegel R. L., Giaquinto A. N., Jemal A., Cancer statistics, 2024, CA: a cancer journal for clinicians, vol. 74, no. 1, pp. 12-49, (2024).
[2] Dinnes J., Deeks J. J., Chuchu N., di Ruffano L. F., Matin R. N., Thomson D. R., Wong K. Y., Aldridge R. B., Abbott R., Fawzy M., Bayliss S. E., Grainge M. J., Takwoingi Y., Davenport C., Godfrey K., Walter F. M., Williams H. C., Dermoscopy, with and without visual inspection, for diagnosing melanoma in adults, Cochrane Database of Systematic Reviews, no. 12, (2018).
[3] Tschandl P., Rosendahl C., Kittler H., The HAM10000 dataset, a large collection of multi-source dermatoscopic images of common pigmented skin lesions, Scientific Data, vol. 5, no. 1, pp. 1-9, (2018).

Speaker

Nikita Zakharov
Samara University
Russia

Discussion

Ask question