A Deep learning-based breast cancer prediction 

It is a comprehensive deep learning framework designed for the automated analysis of breast ultrasound and mammography images. The system uses a hybrid CNN-VGG19 model with transfer learning to achieve high accuracy in detecting and classifying breast lesions as benign, malignant, or normal, thereby assisting radiologists with faster and more reliable diagnoses.

 The above images are related to breast cancer prediction, showing that how many women are suffering yearly.

Breast cancer is one of the most prevalent and life-threatening diseases worldwide. Traditional diagnostic methods, such as mammograms and ultrasounds, rely heavily on manual analysis of medical images, which can be time-consuming and prone to human error. This highlights the need for advanced computational techniques to improve early detection. Deep learning, and specifically convolutional neural networks (CNNs), have emerged as powerful tools for automating medical image analysis, offering high accuracy and efficiency in detecting malignancies.

 Test result images are provided below

I was the primary researcher and developer for the project "A DEEP LEARNING BASED BREAST CANCER PREDICTION". I was responsible for the entire project, from initial concept to implementation and testing. I worked under the guidance of Prof. Bhulakshmi Bonthu, who provided constructive input and innovative ideas for the project's completion. I also had to work with medical professionals like radiologists and oncologists to validate the AI predictions and get feedback to improve the model.

Project Outcome:

The project successfully developed a hybrid CNN-VGG19 model that significantly enhanced breast cancer diagnosis. The model achieved a high accuracy of

97.3% and a sensitivity of 97.1% for detecting malignant cases. It also demonstrated robust performance across various datasets. A key outcome was the model's ability to provide explainable results through visualizations like Grad-CAM heatmaps, which help clinicians trust the system's predictions.

The project's system was designed to integrate into clinical workflows, potentially reducing diagnosis time from 20 minutes to under 2 minutes per case without compromising accuracy. The work fulfilled its objective of creating a reliable decision-support tool that addresses the limitations of traditional, manual diagnostic methods.