background class (bg) in faster rcnn

[Qcatbot]

Can someone tell me why we have to include "bg" class? I am working on an medical image dataset. I am not sure whether I have to include some "bg" class in my training dataset.
This is what I thought to do...
!python train_frcnn.py --network resnet50 -o simple -p train.txt
my train train.txt as follow....
path/1.png, xmin, ymin, xmax, ymax, benign
path/2.png, xmin, ymin, xmax, ymax, malignant
path/3.png, xmin, ymin, xmax, ymax, bg <----------- should I randomly include "bg" class??
I am sorry it is not very clear to me... I searched all over the internet but dont find any useful info about it. Much appreciate if someone could clarify it..
for example...
I randomly extracted the coordinates of a background and included in my train.txt file... is it a right approach? Please see the image below.. yellow box is the tumor and the blue is the randomly picked background...