Improved Cervical Cancer Classification using Generative AI and Language Models

Reza Gheibi, Dean Hougen

Abstract: Cervical cancer continues to be a major global health challenge, and early detection is crucial to improve patient outcomes. In this study, we propose an innovative approach that uses a combination of Generative AI and Large and Small Language Models (LLMs and SLMs) to improve cervical cancer classification and patient interaction. Our methodology involves transforming tabular patient data into textual descriptions, enabling the use of Natural Language Processing (NLP) techniques for classification. We employ generative AI for data augmentation, improving the robustness of predictive models. Additionally, we fine-tune various language models, including text classification transformers from the BERT family and specialized medical models such as PubMedBERT and BioBERT, alongside general-purpose models such as Mistral, LlaMA and GPT-2. We also evaluate the performance of the zero-shot and few-shot learning paradigms of larger language models GPt3&4. Experimental results demonstrate that our approach significantly improves the accuracy of cervical cancer classification compared to traditional methods. This work highlights the potential of integrating generative AI and small language models to enhance medical diagnosis, offering a promising direction for patient-friendly AI-assisted healthcare applications.