Comparing Forward-Forward and Backpropagation in U-Net for Melanoma Image Classification

In recent years, deep neural networks have become essential in medical imaging, especially for precise diagnostic applications. This paper compares two main learning methods for neural networks—Forward-Forward and Backpropagation—focused on the U-Net architecture for classifying dermatological images, specifcally melanomas. The Forward-Forward approach, which sidesteps traditional gradient-based Backpropagation in favor of a simpler, unidirectional process, offers a more computationally effcient alternative. In contrast, Backpropagation is a well-established method for optimizing network weights, especially for complex tasks where high accuracy is crucial. We trained U-Net models on a dataset of melanoma images, evaluating both their computational and diagnostic performance. The fndings show that while Backpropagation achieves higher accuracy and precision, the Forward-Forward method stands out in computational effciency, making it valuable in resourcelimited settings. This study highlights the balance between computational speed and diagnostic accuracy, suggesting potential ways to optimize neural networks for medical diagnostics.