How to identify structures in a 3D matrix/image?

I need to identify all the structures/volumes of a 3D matrix. The values of the matrix represent energy that span more many orders of magnitude (7+).

What I did is to plot the histogram of every pixel in order to identify the background value. Then I created a binary matrix of equal size, in which the zero values represent the values below the background, and 1's denote the values above this level.

Then, I use the "bwconncomp" function to identify and then label the clusters of points.

However, despite the many identified features, the method identifies also a very large feature, which is not realistic because when searching for its local maximum you minde many of those. In addition, the nature of the data suggests that there should be many smaller events and not one huge unified.

Therefore, I need a method that chops this large object or an improvement of the method just mentioned to resolve this large feature.

Any suggestions? C