IMPLEMENTATION OF A CONVOLUTIONAL NEURAL NETWORK APPLIED TO THE PROCESS OF TIME-LAPSE MEDICAL IMAGING

This paper presents the implementation of a convolutional neural network (CNN) applied to the prediction of values derived from time series of segmented medical images, according to a methodological flow defined by a patent on time-lapse imaging. The approach involves image acquisition, spatiotemporal alignment, segmentation of the region of interest, and extraction of morphometric features, including area, perimeter, density, and entropy. These characteristics are organized into temporal sequences and used as input to the predictive model. The training is carried out with segmented images, reserving the last of each sequence for validation. The model outputs are converted into visual representations, allowing comparison with the actual images. The evaluation uses MAE and RMSE metrics in original and binarized images. The system is inserted in the context of medical informatics, with application in the monitoring of tumor structures in sequential examinations over time.