Abstract:
Cancer represents an abnormal proliferation of cells within the body tissue, capable of rapid dissemination throughout various regions. Among the most prevalent cancer types, breast cancer stands out as a significant health concern. It exhibits uncontrolled growth within the breast tissue, leading to numerous fatalities, often due to delayed diagnosis and inadequate treatment. Accurate detection and classification of breast cancer have posed significant challenges, primarily owing to the massive volume of images requiring analysis due to escalating cancer cases. The presence of similar tissue characteristics and cluster formations within a size range of 0.05mm – 1mm further complicates localization and reduces classification accuracy. Although machine learning algorithms like support vector machines and decision trees have modestly improved detection accuracy, they still rely on feature extraction from images, leading to inefficiencies.
To address these limitations, this research focuses on utilizing a state-of-the-art approach in medical image analysis, specifically a modified Convolutional Neural Network (CNN) technique, capable of autonomously learning features and making precise predictions. By adopting transfer learning, we leverage a pre-trained model (AlexNet architecture) and tailor it to our breast cancer classification task. The method incorporates reflection and rotation as augmentation techniques, enhancing the dataset to train the model effectively. The dataset undergoes preprocessing to improve the quality of the images before feeding them into the model for training.
The proposed model achieved impressive performance on the test dataset, with accuracy, sensitivity, specificity, precision, and F1 score reaching 95.80%, 95.00%, 80.00%, 92.30%, and 93.63%, respectively. These results demonstrate significant enhancements in classification accuracy compared to existing literature utilizing deep learning techniques and the MIAS breast cancer dataset. Implementing this model can aid medical professionals in making accurate and timely classifications, effectively reducing the time wasted on manual breast cancer analysis.
Enhancing Breast Cancer Classification with a Modified Convolutional Neural Network Technique.  GET MORE, ACTUARIAL SCIENCE PROJECT TOPICS AND MATERIALS
