A hybrid parallel convolutional spiking neural network for enhanced skin cancer detection.

The most widespread kind of cancer, affecting millions of lives is skin cancer. When the condition of illness worsens, the chance of survival is reduced, and thus detection of skin cancer is extremely difficult. Hence, this paper introduces a new model, known as Parallel Convolutional Spiking Neural Network (PCSN-Net) for detecting skin cancer. Initially, the input skin cancer image is pre-processed by employing Medav filter to eradicate the noise in image. Next, affected region is segmented by utilizing DeepSegNet, which is formed by integrating SegNet and Deep joint segmentation, where RV coefficient is used to fuse the outputs. Here, the segmented image is then augmented by including process, such as geometric transformation, colorspace transformation, mixing images Pixel averaging (mixup), and overlaying crops (CutMix). Then textural, statistical, Discrete Wavelet Transform (DWT) based Local Direction Pattern (LDP) with entropy, and Local Normal Derivative Pattern (LNDP) features are mined. Finally, skin cancer detection is executed using PCSN-Net, which is formed by fusing Parallel Convolutional Neural Network (PCNN) and Deep Spiking Neural Network (DSNN). In this work, the suggested PCSN-Net system shows high accuracy and reliability in identifying skin cancer. The experimental findings suggest that PCSN-Net has an accuracy of 95.7%, a sensitivity of 94.7%, and a specificity of 92.6%. These parameters demonstrate the model's capacity to discriminate among malignant and benign skin lesions properly. Furthermore, the system has a false positive rate (FPR) of 10.7% and a positive predictive value (PPV) of 90.8%, demonstrating its capacity to reduce wrong diagnosis while prioritizing true positive instances. PCSN-Net outperforms various complex algorithms, including EfficientNet, DenseNet, and Inception-ResNet-V2, despite preserving effective training and inference times. The results obtained show the feasibility of the model for real-time clinical use, strengthening its capacity for quick and accurate skin cancer detection.