Histopathological Image based Skin Cancer Classification Using CNN

Published: 6 December 2021| Version 1 | DOI: 10.17632/d48b5zybck.1


The dataset consisting of H&E-stained digital microscopic histology images which was collected from 354 cancer patients on A.H.R.I. The dataset has four Classes (Benign,BCC,SCC and Melanoma). The dataset is divided into benign and malignant tumor of biopsies. Small patches were extracted at four magnifications of ×40, ×100, ×200, and ×400. The benign tumors were classified into eight subclasses which were Melanocytic Nevus, Prome, Pilomatrixomas, Sebonlaric Keratosis (Pigmented), Trico blastoma, Verica, Dermatofibroma, Mull scum Contagiosum, Infundibular Cyst and the malignant tumors were also classified into three subclasses which were basal cell carcinoma (BCC), squamous cell carcinoma (SCC) and Melanoma sometimes called Malignant Melanoma (MM). The dataset contains 16,099 total histology images with data augmentation and without data augmentation 4,357 total histology images. The owner of this Dataset is Alemu Abate Asheber MSc Software Engineer) . There is no third party in this dataset.


Steps to reproduce

The Olympus BX63 Digital Motorized Upright Advanced microscope camera configuration with different magnification objects i.e. 4x,10x,20x and 40x, to capture malignant skin cancer histopathology images with the class of Malignant BCC, Malignant SCC, Malignant Melanoma and Benign are channel count =3, Frame count=1,type= 48bit RGB color, color space= sRGBIEC61966-2.1,size (pixel) =1600x1200, size (calibration) =17.9mmx11.mm, calibration(x) = 11.162 µm / pixel , calibration (y) = 9.302 µm / pixel, origin (x) = 0µm, origin (y) = 0µm, total magnification = 0.63x, contrast= 2.00, exposure time= 1.21ms, binning=1x1,gain=1.00x,gamma= 1.00, sharpness=2.


Addis Ababa Science and Technology University


Image Database