In general, fine roots in plants are defined as roots that are less than 2 mm in diameter that grow and die in a relatively short period of time. Fine root production can account for approximately 30% of net primary production of a forest. Therefore, fine root dynamics significantly affect carbon cycling of a forest. Fine root dynamics can be analyzed by extracting fine root area from soil profile time series images. However, identifying fine roots from these images is time-consuming. Recently, use of deep learning-based software, ARATA, has been proposed as an efficient method to automatically extract fine roots from soil profile images. The root dynamics estimated from the automatic extraction of fine roots by ARATA showed positive correlations with those of the manual extraction. The ARATA successfully extracted 50-90% of the number of growing and dying roots that were identified by manual extraction. However, several challenges remain in the use of ARATA for extracting fine root dynamics. One of them is to detect and remove the noises in the soil profile images that are extracted as roots by ARATA. Another challenge is that it is still unknown whether ARATA can extract growing roots accurately in terms of area. Therefore, we aimed to assess and improve the extraction of fine root dynamics using ARATA in this study. This study has two objectives: 1) to remove noise from growing and dying root images; and 2) to evaluate the root growth area extracted by ARATA by comparing them with those of manual extraction. We developed a method that removes noises from ARATA extracted images by combining multiple image processing techniques such as the Hough transform. The method successfully removed most of the noises from the extracted images. We then evaluated the area of root growth in the noise-removed images. We found ARATA extracted up to approximately 80% of the area of fine root growth.