AAI_2025_Capstone_Chronicles_Combined

Breast Tumor Classification Using Quantum Neural Networks

18

to the same bit but a different target were removed. Once all this down sampling is taken into

account, the model performance doesn’t sound that terrible at 80%.

The most significant boost in performance came from exploring different quantum

feature mappings. The initial quantum feature mapping simply mapped the classical bits to

|0> and |1> , the quantum equivalent of binary. However, qubits can also exist in a

superposition of those states. After normalizing the features to values between 0 and 1, the

middle values were put into a superposition of |0> and |1> , effectively having each feature in

one of 3 states: |0> , |1> , and an equal super position of |0> and |1> (similar to Figure 7).

This relatively simple change resulted in the model accuracy of over 90% with just 10 features

selected. The classical equivalent would be to do the classification with only 10 bits of

information. If this work is stored as Unicode, a single letter would require more bits than this.

This illustrates the potential of quantum computing combined with AI: successfully predicting

cancer with less information that is used to store a single letter.

271