Anki0909 / BreakHist-Dataset-Image-Classification

BreakHist Dataset contains histopathological images of eight types of breast cancer, including four benign cancer and for malignant cancer. In this project, I have trained and fined tuned many of the existing CNN models to get over 80% accuracy in multi-class classification.

The tables below gives accuracy of each model for each magnification zoom presents in the dataset upto three decimal units. The values in brackets are F1 score

CNN models with FCN at the end

Magnification/CNN Model ->	VGG-16	VGG-19	Xception	Resnet	Inception	Inception-Resnet-V3
40X	0.802 (0.803)	0.652 (0.685)	0.831 (0.831)	0.859 (0.858)	0.853 (0.858)	0.818 (0.813)
100X	0.867 (0.877)	0.709 (0.708)	0.786 (0.794)	0.911 (0.917)	0.834 (0.827)	0.845 (0.837)
200X	0.841 (0.839)	0.749 (0.756)	0.812 (0.813)	0.857 (0.853)	0.799 (0.806)	0.854 (0.859)
400X	0.871 (0.869)	0.799 (0.799)	0.761 (0.758)	0.903 (0.907)	0.799 (0.796)	0.842 (0.844)

CV score on Logistic Regression Model trained on features extracted from CNN models

Magnification/CNN Model ->	VGG-16	VGG-19	Xception	Resnet	Inception	Inception-Resnet-V3
40X	0.685 (0.675)	0.565 (0.547)	0.858 (0.856)	0.908 (0.906)	0.839 (0.836)	0.854 (0.850)
100X	0.732 (0.725)	0.633 (0.623)	0.840 (0.837)	0.902 (0.900)	0.826 (0.822)	0.863 (0.862)
200X	0.864 (0.862)	0.725 (0.718)	0.940 (0.954)	0.959 (0.958)	0.919 (0.917)	0.961 (0.960)
400X	0.952 (0.952)	0.876 (0.874)	0.982 (0.982)	0.983 (0.983)	0.983 (0.983)	0.982 (0.982)

CV score on Linear Support Vector Machine Model trained on features extracted from CNN models

Magnification/CNN Model ->	VGG-16	VGG-19	Xception	Resnet	Inception	Inception-Resnet-V3
40X	0.644 (0.640)	0.543 (0.530)	0.857 (0.856)	0.905 (0.905)	0.855 (0.853)	0.851 (0.849)
100X	0.711 (0.704)	0.603 (0.595)	0.830 (0.829)	0.895 (0.894)	0.826 (0.822)	0.864 (0.863)
200X	0.848 (0.847)	0.700 (0.693)	0.943 (0.942)	0.961 (0.961)	0.916 (0.916)	0.958 (0.958)
400X	0.950 (0.949)	0.868 (0.867)	0.983 (0.983)	0.983 (0.983)	0.983 (0.983)	0.980 (0.980)