Classification on Digital Pathological Images of Breast Cancer Based on Deep Features of Different Levels

Figure 5

Experimental structure. Data 1 are obtained by Sliding window slicing. Data 2 are obtained by Random window slicing. Data 3 are obtained by Sliding + Random window slicing. Data 4 are obtained by Sliding + Class Balance Random window slicing. Data N are obtained by the best slicing strategy. The orange region represents the effective feature extraction region. The gray region represents the invalid feature extraction region. GMP represents the global max pooling layer. MLCS represents the abbreviation of ML classifier set.