Deep integrated pipeline of segmentation guided classification of breast cancer from ultrasound images

Breast cancer has become a symbol of tremendous concern in the modern world, as it is one major causes mortality worldwide. In this regard, breast ultrasonography images are frequently utilized by doctors to diagnose at an early stage. However, complex artifacts and heavily noised make diagnosis great challenge. Furthermore, ever-increasing number patients being screened for necessitates use automated end-to-end technology highly accurate low cost short time. concern, develop integrated pipeline image classification, we conducted exhaustive analysis preprocessing methods such K Means++ SLIC, well four transfer learning models VGG16, VGG19, DenseNet121, ResNet50. With Dice-coefficient score 63.4 segmentation stage accuracy F1-Score (Benign) 73.72 percent 78.92 classification stage, combination UNET, VGG16 outperformed all other combinations. Finally, have proposed end pipelining framework which includes with SLIC capture super-pixel features from artifact images, complementing semantic modified U-Net, leading tumor using approach pre-trained densely connected neural network. The can be effectively implemented assist medical practitioners making more timely diagnoses cancer.