Mitigating Algorithmic Bias in Cancer Site Classification Models

Puntos clave

• This research aims to assess how HiSAN-generated embeddings can reduce algorithmic bias in cancer classification models.
• Utilized SEER data for generating HiSAN embeddings.
• Applied post-training pruning as an audit method.
• Examined the impact of demographic factors on cancer classification outcomes.
• HiSAN embeddings were shown to classify cancer sites effectively.
• Significant reductions in algorithmic bias were observed after using pruning techniques.