Deep Learning with DenseNet121: Architecture and Application to Breast MRI Classification

It reviews the fundamental components of neural networks, including neurons, weighted connections, activation functions, softmax classification, cross-entropy loss, gradient descent, backpropagation, learning rate, and optimization.</p><p dir="ltr">The presentation then provides a step-by-step examination of a 3D DenseNet121 architecture designed for volumetric breast MRI.

It explains convolutional feature extraction, batch normalization, ReLU activation, pooling, dense connectivity, transition layers, feature-map growth and compression, and global average pooling. Particular emphasis is placed on how DenseNet reuses features by connecting each layer to preceding layers and progressively transforms a 3D MRI volume into a compact feature representation for classification.</p><p dir="ltr">The material uses the distinction between inflammatory breast cancer (IBC) and non-inflammatory locally advanced breast cancer (LABC) as the motivating clinical application, connecting fundamental deep learning concepts to a real-world medical imaging workflow.</p>