A Pyramid Architecture-Based Deep Learning Framework for Breast Cancer Detection

Algorithm 1

Tissue-level pathological RoI extraction.

Input: WSI image I, patch level DeconvNet for scale l, C_l, layer number L, and confidence threshold t.

Output: Selected patches Ps.

1: Generate patches Ps_L-1 with step w and h in I_L-1, and location code LCIS_L-1.

2: patches initialization with Ps = Ps_L-1.

3: for i = L-2 to 2 do

4: if Ps is empty then

5:  Break

6: for patch p in Ps do

7:  Calculate cancer confidence of p named c with C_i

8:  if c > t then

9:   Add LCI_p to LCIS_i

10: for LCI in LCIS_ido

11:  Calculate LCI in i-1 layer

12: Generate patches with all LCI_s named Ps_i-1

13: Set current patch set Ps = Ps_i-1