A Heuristic Neural Network Structure Relying on Fuzzy Logic for Images Scoring

Traditional deep learning methods are suboptimal in classifying ambiguity features, which often arise noisy and hard to predict categories, especially, distinguish semantic scoring. Semantic scoring, depending on logic implement evaluation, inevitably contains fuzzy description misses some concepts, for example, the ambiguous relationship between normal probably always presents unclear boundaries (normal-more likely normal-probably normal). Thus, human error is common when annotating images. Differing from existing that focus modifying kernel structure of neural networks, this article proposes a dominant fully connected layer (FFCL) breast imaging reporting data system (BI-RADS) scoring validates universality proposed structure. This model aims develop complementary properties paradigms, while constructing rules based analyzing thought patterns, particularly reduce influence conglutination. Specifically, semantic-sensitive defuzzifier projects features occupied by relative categories into space, decoder modifies probabilities last output referring global trend. Moreover, space two shrinks during phases, as positive negative growth trends one category appearing among its relatives were considered. We first used Euclidean distance zoom real scores predicted scores, then employed sample t test method evidence advantage FFCL architecture. Extensive experimental results performed curated subset digital database screening mammography dataset show our can achieve superior performances both triple multiclass classification BI-RADS outperforming state-of-the-art methods.