Efficient Deep Learning Architecture for Detection and Recognition of Thyroid Nodules

Figure 6

Different kernel sizes corresponding to different dilation rates (here, the 3 × 3 convolution kernel size is considered as an example): (a) the dilation rate is 1(conventional convolution kernel); (b) the dilation rate is 2; (c) the dilation rate is 3; (d) the dilation rate is 4.

(a)

(b)

(c)

(d)