The choice of activation functions is central to the neural network (NN) training process. Recent years have seen increased focus within the deep learning community on developing efficient, high-performance activation functions. In this work, we use a new family of activation functions parameterized by hypergeometric functions. This family incorporates multiple trainable parameters, enabling the NN to adapt a variety of activation functions during training. We focus on Bessel functions of the first kind \(J_\nu \) , representing a subset within this broader family of hypergeometric-based activations. This paper presents preliminary results on the performance of Bessel-type activation functions for clinical image classification, comparing their performance to the widely used ReLU activation function.