Attention-Based Deep Learning Approach for Breast Cancer Histopathological Image Multi-Classification

. 2024 Jul 1;14(13):1402. doi: 10.3390/diagnostics14131402

Algorithm 1 Constructing an oversampled balanced dataset

Input:
$m a i n_d i r$ : Directory containing class folders with images
Output:
Oversampled balanced dataset
1:
Identify and enumerate all classes in $m a i n_d i r$
2:
Count the number of images per class
3:
$m a x_c l a s s_c o u n t \leftarrow$ maximum image count across classes
4:
$o v e r s a m p l e_t a r g e t \leftarrow 3 \times m a x_c l a s s_c o u n t$
5:
for each class c do
6:
$i m a g e_c o u n t \leftarrow$ count of images in class c
7:
if $i m a g e_c o u n t < o v e r s a m p l e_t a r g e t$ then
8:
Oversample class c up to $o v e r s a m p l e_t a r g e t$ using $t r a n s f o r m$
9:
end if
10:
for $i \leftarrow 1$ to $o v e r s a m p l e_t a r g e t$ do
11:
$i n d e x \leftarrow (i - 1) mod i m a g e_c o u n t$
12:
$i m a g e_p a t h \leftarrow$ path of $i n d e x$ -th image in class c
13:
Load image from $i m a g e_p a t h$
14:
Apply $p r e p r o c e s s$ to the image
15:
if $i m a g e_c o u n t < m a x_c l a s s_c o u n t$ then
16:
Apply $b a l a n c e$ to augment the image
17:
end if
18:
Apply $t r a n s f o r m$ to the image
19:
Return image and class label
20:
end for
21:
end for