Abstract:
Breast cancer diagnosis using histopathology images and advanced deep learning
techniques. The study evaluates five models: MobileNetV2, ResNet50V2, DenseNet121,
DenseNet201, and a custom CNN. Among these, DenseNet201 emerged as the most
effective model, achieving a validation accuracy of 91.48% and a validation loss of 0.2134,
showcasing its robust performance and superior generalization capabilities. Similarly,
DenseNet121 demonstrated strong results with a validation accuracy of 88.92% and a
validation loss of 0.3013, making it another reliable option for classification tasks.While
ResNet50V2 exhibited the highest training accuracy of 97.64%, its validation accuracy of
82.66% highlighted the need for further fine-tuning to address potential overfitting.
MobileNetV2 achieved a validation accuracy of 80.42%, emphasizing its efficiency in
training with a low training loss of 0.1238 but limited generalization compared to the
DenseNet models. The custom CNN achieved a validation accuracy of 87.26%, proving its
capability as a lightweight alternative suitable for deployment in resource-constrained
environments.The study also includes ethical considerations such as ensuring patient data
privacy and equitable access to AI-driven diagnostics. Furthermore, a sustainability plan
was devised to minimize environmental impact through energy-efficient practices and
telemedicine applications. A user-friendly web application was developed to enable
healthcare providers to access the trained models, facilitating their integration into clinical
workflows. This research significantly advances breast cancer diagnostics, offering
scalable, ethical, and accessible AI solutions to improve patient care and healthcare
efficiency.
