In recent years, gradual patterns have gained the attention of the data science community and several algorithms have been designed to extract these patterns from different data models. On some data, like imprecise data, one of the biases in traditional algorithms is that they define graduality as an increase/decrease in value. Therefore, some of the graduations extracted are just a noise effect in the data. To remedy this problem, this paper proposes a method which introduces into the mining process, a gradual threshold from which to consider a graduality. Experiments on different databases show that our proposal reduces computation times and the number of patterns generated by focusing on patterns of interest. Additionally, she extracts gradual patterns in some cases where traditional approaches fail.