AAI_2025_Capstone_Chronicles_Combined

Breast Tumor Classification Using Quantum Neural Networks

11

classic text by Nielsen and Chuang (Nielsen & Chuang, 2000) or the more compact text by

Hirvensalo (Hirvensalo, 2004).

Quantum Neural Networks

Quantum neural networks are similar to classical neural networks in that there is an input

layer, hidden layer(s), and finally the output layer. Both the input and output are classical, i.e.

traditional bits and bytes. The key difference is that the input is mapped to a quantum state,

which then goes through the quantum neural network (consisting of quantum circuits), which is

then collapsed through measurement to obtain the prediction output. Figure 8 details this flow.

An advantage of quantum neural networks is that they are resistant to overfitting and useful

when there is limited training data (Hirai, 2024).

Figure 8

Diagram of a quantum neural network, based on (Karthinkeyan, Akila, Sumathi, & Poongodi, 2025) and some images from (Purkeypile, 2009) .

An example this project was inspired by was using quantum neural networks on the

MNIST handwritten images data set (LeCun, Cortes, & Burges, 2024) and required the images

to be scaled down to 4x4 black and white and was a binary classification exercise- so only 2 of

the 10 digits were selected at a time. These selected images were then mapped into 16 (4x4)

qubits, with 0 mapping to the quantum state |0> and 1 mapping to the quantum state |1> . This

example is covered in detail in (Pattanayak, 2021). Unsurprisingly, scaling down the images to

this degree had a significant impact on the performance of binary classification. This example

264