EM cell segmentation, segment dark area with specific color in presence of noise and other cells.

I am working on electron microscopic Images and I must segment mitochondria cell and connection between them, I used Manual Threshold and Otsu's method but the problem is in images we had lots of other cells and noise . I tried to consider regions and areas with specific size but again I can not get rid of unnecessary areas . The original images is like this :

and the darker area (manually segmented) that I must keep in my work are like :

The output of Manual thresholding and Otsu's method is like :

I can use codes like :

BW1 = bwareaopen(bwp, 1500,4);
[Label,Total]=bwlabel(BW1,8);
display(Total);
s = regionprops(BW1,originalImage,{'MeanIntensity'});

or s = regionprops(BW1,'basic');

to find objects based on their area and size but the unwanted object will remain in image . based on the thresholding result (last image) how can I keep objects with darker filled inside or if I crop same areas from original image how can I keep them based on their properties. ?