medical processing

The classification of Doppler ultrasound images

is very important for conception prediction. However it is a

challenging problem that suffers from a variable length of those

images with a dimension gap between them. In this study, we

propose a latent representation weight learning method (LRWL)

for conception prediction with Doppler ultrasound images. Unlike

most existing related methods, LRWL can process a variable

length of multiple images, particularly with an irregular multi-image issue. LRWL can extract the relation between the images

and then learn the latent representative weight of each image.

Comparatively, we also propose another method—spatiotemporal

interaction measurement (SIM), to validate the experimental

assumption that LRWL can describe the role of each image more

accurately. Then we integrate the images with the weights and

the diagnostic indices, as the input data to a deep learning (DL)

model to predict the successful conception. Finally, we conduct

the comprehensive experiments with the classification tasks on

the real irregular reproduction datasets and two other synthetic

regular datasets. Experimental results show that the proposed

LRWL outperforms existing relative methods and is suitable for

the irregular multi-image datasets. It can be efficiently optimized

under the limited-memory Broyden-Fletcher-Goldfarb-Shanno

bound constraint algorithm (L-BFGS-B), and the alternating

direction minimization (ADM) framework. Thus, the proposed

method can achieve the good performance with high accuracy

and good convergence.