A novel bio-inspired fibonacci-based convolutional neural network for accurate skin disease diagnosis using private dataset

Abstract

Early detection of cancers in general, and skin cancer and dermatological diseases in particular, plays a crucial role in recovery and improving treatment outcomes. Early disease detection is one of the most significant global health challenges. Recently, advancements in computing, especially artificial intelligence, have been harnessed for the early prediction of diseases, including dermatological ones. Deep learning convolutional neural networks have demonstrated high performance in analyzing medical images, but balancing high diagnostic performance with computational efficiency has been one of the major challenges this technology has faced. In this study, the Fibonacci sequence and the golden ratio (1.618) were used to design a convolutional neural network architecture to improve the diagnosis of multicategory skin lesions. One of the most important features of the resulting convolutional neural network architecture is its unique layer configuration. A dataset of realistic skin images was used to test and evaluate the proposed model and compare the results with the latest convolutional neural network architectures, including MobileNetV2, ResNet50 and VGG16. The proposed Fib-CNN model in this study achieved satisfactory performance with an accuracy of 90.8%, a precision of 0.907, a recall of 0.911, and an ROC-AUC value of 0.958, which confirms the superiority of the Fib-CNN model over the basic models in terms of diagnostic accuracy and also reducing computational complexity by up to 22%. The results demonstrate the effectiveness of the Fib-CNN model in diagnosing skin diseases by improving the depth-to-width ratio of convolutional layers, reducing background interference, and generating more accurate focus maps of skin lesions.