Breast Cancer is one of the top leading common cancer among women and less frequently in men and it is marked as one of the major causes of cancer-related mortalities. This proposed research adapts the merits of an advanced Computer Aided Diagnosis (CAD) tool which has the potentiality of automatic identification of mass regions and finding the abnormalities on tubes, lumps and ducts among ultrasound images. This research presents a deep learning-based approach using Improved SqueezeNet model combined with Adam& Dingo optimizer to enhance the degree of classification accuracy. This proposed scheme follows a pipeline process for diagnosis of breast cancer that involves: Data Collection, Noise Reduction & Preprocessing, Feature Extraction, Hyper parameterization using Adam & Dingo Optimizer and Evaluation. As an initial step, Breast Ultra Sound Image dataset is collected from Kaggle. As a next step, a Median Filter with Contrast-Limited Adaptive Histogram Equalization (CLAHE) techniques are incorporated for eliminating the noise from the ultrasound images and to enhance the contrast on mass regions for accurate detection. Then, an improved SqueezeNet model is incorporated for the purpose of feature extraction and the extracted features undergoes Hyper parameterization utilizing a Hybrid combination of Dingo-Adam optimizer for finetuning. Simulations of the method are applied to the Breast Ultra Sound Images Dataset and the experimental results of proposed Improved SqueezeNet model exhibits superior results when compared with other deep learning models like VGG16, MobileNetV2 and DenseNet121.