Dual-modality synthetic mammogram construction for breast lesion detection using U-DARTS

Multimodal image fusion is an emergent research area for cancer detection. It provides a wide variety of visual qualities for the accurate medical diagnosis. However, this process requires accurate registration of each image modality for its efficient and effective use. To address the aforementioned issue, a novel synthetic mammogram construction model is proposed. An image enhancement approach is applied to enhance the image quality. Dual-modality structural feature (DMSF) based mapping function is designed to transform a mammogram from a thermal image segment. This paper also proposes modified Differentiable ARchiTecture Search (DARTS) named as (U-DARTS) to detect and classify the breast lesion. In U-DARTS, a stochastic gradient descent optimizer is used. The proposed approach is tested over DMR and INbreast datasets. The results exhibit a significant improvement in the performance of the proposed model over the existing techniques. The validation and testing accuracies of 98% and 91%, respectively, are achieved. Overall, the proposed approach establishes supremacy in so far as the mammogram construction and further lesion detection are concerned.