IMAGE PROCESSING FOR DETECTING MELANOMA SKIN CANCER USING AN OPTIMIZED ROTATION–INVARIANT COORDINATE CONVOLUTIONAL NEURAL NETWORK

Cancer is a dangerous disease caused by the unregulated proliferation of bodily cells. Cancer can grow in any part of the human body and may contain billions of cells. Skin cancer is a common type of cancer that originates on the top layer of the skin. Melanoma is one category of skin cancer, this is the most dangerous disease and leads to mortality. Generally, the machine learning approaches are mostly employed to diagnose skin cancer using protein sequences and other forms of imaging. But some disadvantages occur while using the machine learning systems, like requiring human-engineered features, difficulty, and being time-consuming. In this manuscript, Image Processing for detecting Melanoma Skin Cancer using an optimized Rotation–Invariant Coordinate Convolutional Neural Network (IP-MSCD-RICCNN) is proposed for skin cancer detection. The input images are gathered through the International Skin Imaging Collaboration (ISIC) dataset. Then the images are pre-processed using the Gillijn De Moor Filter (GDMF) to remove the noise. The pre-processed image is given to the Generalized Intuitionistic Fuzzy C-Means Clustering (GIFCMC) for segmenting the Region of Interest (ROI) from the skin image. The segmented image is supplied to the Newton Time-Extracting Wavelet Transform (NTEWT) to extract Gray Level Co-occurrence Matrix (GLCM) features such as Contrast, Correlation, Entropy, and Inverse. After feature extraction, the Rotation–Invariant Coordinate Convolutional Neural Network (RICCNN) is used to classify the skin images as malignant melanoma and benign melanoma. RICCNN generally does not adapt any optimization approaches to determine the optimal parameters for exact prediction. Therefore, the Mutated Coati Optimization Algorithm (MCOA) is employed to enhance the RICCNN for skin image classification. The proposed IP-MSCD-RICCNN method is implemented in Python. The proposed method achieves 26.36%, 20.69%, and 35.29% higher accuracy, 32.85%, 19.63%, and 25.47% higher precision, and 16.57%, 21.47%, and 30.56% greater sensitivity when compared to the existing models: skin cancer diagnosis from dermoscopy imageries utilizing deep learning with fuzzy k-means clustering (DI-SCD-DL-FKC), automatic detecting skin cancer by consolidating texture features with convolutional neural networks in dermoscopy images (DI-SCD-CNN-TF), and detection of skin cancer utilizing combined decision of deep learners (SCD-CD-DCNN) respectively.