Skin Lesion Synthesis and Classification Using an Improved DCGAN Classifier

The prognosis for patients with skin cancer improves regular screening and checkups. Unfortunately, many people do not receive a diagnosis until the disease has advanced beyond point of effective therapy. Early detection is critical, automated diagnostic technologies like dermoscopy, an imaging device that detects lesions early in disease, are driving factor. lack annotated data class-imbalance datasets makes using methods challenging lesion classification. In recent years, deep learning models have performed well medical diagnosis. such require substantial amount training. Applying augmentation method based on generative adversarial networks (GANs) to classify plausible solution by generating synthetic images address problem. This article proposes synthesis classification model Improved Deep Convolutional Generative Adversarial Network (DCGAN). proposed system generates realistic several convolutional neural networks, making training easier. Scaling, normalization, sharpening, color transformation, median filters enhance image details during uses generator discriminator global average pooling 2 × fractional-stride, backpropagation constant rate 0.01 instead 0.0002, most hyperparameters optimization efficiently generate high-quality images. As classification, final layer Discriminator labeled as classifier predicting target class. study deals binary two classes—benign malignant—in ISIC2017 dataset: accuracy, recall, precision, F1-score performance. BAS measures accuracy imbalanced datasets. DCGAN Classifier demonstrated superior performance notable 99.38% 99% F1 score, BAS, outperforming state-of-the-art models. These results show can accurately them, it promising tool learning-based analysis.