Skin Cancer Image Detection and Classification by CNN based Ensemble Learning

Abstract

Melanoma is accounted as a rare skin cancer responsible for a huge mortality rate. However, various imaging tests can be used to detect the metastatic spread of disease with a primary diagnosis or on clinical suspicion. Focus on melanoma detection, irrespective of its unusual occurrence, is that it is often misdiagnosed for other skin malignancies leading to medical negligence. Sometimes melanoma is detected only when the metastasis has entered the bloodstream or lymph nodes. So, effective computational strategies for early detection of melanoma are essential. There are four principal types of skin melanoma with two sub types: Superficial spreading, nodular, lentigo, lentigo maligna, Acral lentiginous, and Subungual melanoma. Amelanotic melanoma, one particular type of melanoma, exists in all kinds of skin tones. Classifications of melanoma with its classes are focused on in this research. The ensemble classifier models, namely Adaboost, random forest, voted ensemble, voted CNN, Boosted SVM, and Boosted GMM, have been used in melanoma classification to address misclassification errors, overfitting issues, and improve accuracy. The results of the ensemble classifier achieve high classification accuracy. However, imbalanced classification is found in all six classes of melanoma. Transfer learning and ensembled transfer learning approaches are implemented to reduce the imbalanced classification issues, and performances are analyzed. Four ML/DL models, six ensembled models, four transfer learning models, and five ensembled transfer learning models are used in this investigation. Implementation of all the 19 classifiers is analyzed using standard performance metrics such as Accuracy, Precision, recall, Mathew's correlation coefficient, Jaccard Index, F1 measure, and Cohen's Kappa.

Performance Improvement of Melanoma Detection Using a Multi-Network System Based on Decision Fusion

“…In the paper's conclusions, the authors note that the Ensemble voted, GMM, and random forest classifiers exhibit comparatively lower performance than the Adaboost and boosted SVM classifiers. Notably, the ensemble CNN achieves an impressive accuracy rate of 98.67% [35]. In conclusion, the field of melanoma detection through CNNs and DL techniques has seen significant exploration over the years, with various methodologies offering promising results.…”

“…In [35], the authors explore the application of various machine-learning techniques to construct a high-performance ensemble classifier for six distinct skin lesions. Specifically, the researchers utilize Adaboost, voted ensemble, random forest, boosted Gaussian Mixture Model (GMM), voted Convolutional Neural Network (CNN), and boosted Support Vector Machine (SVM).…”

“…As stated in reference [36], the complete ISIC database comprises a total of 93,083 images, with 71,372 designated for training purposes and 21,711 earmarked for testing, constituting the most extensive publicly available collection of dermoscopic skin lesion images [35]. In our previous study conducted in 2022 [37], we demonstrated that ISIC is the preferred database among researchers for developing high-performance melanoma detection systems.…”

An Extensive Investigation into the Use of Machine Learning Tools and Deep Neural Networks for the Recognition of Skin Cancer: Challenges, Future Directions, and a Comprehensive Review

“…Dermoscopic images are classified into seven sub-types in other studies related to the categorization of skin cancer. Random forests have been used to implement this; we mention the works [26,27]. In this investigation, the process for creating a random forest is somehow slightly different.…”