AAI_2025_Capstone_Chronicles_Combined

Breast Tumor Classification Using Quantum Neural Networks

17

Figure 12

Performance of initial model, based on the MNIST example in (Pattanayak, 2021)

The original quantum neural network attempted consisted of a single hidden layer with a

(quantum) node per feature. There were a few minor changes that had to be made to work with

this breast cancer data set. The selected features in the data set were represented by a 1

dimensional array. This quantum neural network also has an additional qubit for the predicted

target (called the readout qubit). Of course, quantum gates are applied to those qubits and can

be seen in Figure 11.

A clear area to explore was increasing the number of hidden quantum layers in the

model. To do this, the code was refactored to allow for an arbitrary number of similar layers. This

showed a remarkable improvement in performance, with approximately 80% accuracy as shown

in Figure 13. On the surface, 80% accuracy doesn’t sound very impressive. However, for

performance reasons the number of features was reduced from 30 to 10. Additionally, those

features were also mapped to a binary 0 or 1 before the quantum feature mapping. This also

meant that some of the data was discarded: any cases where the selected features all mapped

270