Feature Extraction Method for Early-Stage Colorectal Cancer Using Dual-Tree Complex Wavelet Packet Transform
Daigo Takano, Teruya Minamoto · 2021
Cancer detection using deep-learning techniques is being actively researched in the recent years. However, the results ob-tained by such methods are difficult to interpret. The proposed feature extraction method applies principal component analysis to the variance of the frequency components obtained from the wavelet transform and employs the top three principal component scores in the contribution ratio as features. This method comprises a linear process, making it easier to infer the basis for decisions than the deep-learning. Classification experiments using the ob-tained features confirmed that the classification performance was better than that of the existing methods. For some images, the val-ues of features differed significantly between early-stage colorectal cancer and normal areas, and the detection of early -stage colorec-tal cancer may be possible without using a learning machine in the future.