Fine-tuned T5-base model for medical report summarization: outperforming extractive methods and the pre-trained Gemini 2.0 flash language model

Within NLP as an important and progressing domain in Computer Science, there is a significant scope of research in the task of generation of precisely well and grounded Summary. Due to the presence of frequently used medical abbreviations and complex and domain-specific terminology, generating summary of medical text reports becomes a challenging task. For an individual with limited medical background, receiving the gist of medical/health conditions through the summary of the complete text report of CT scan or MRI, before approaching the medical practitioner can be advantageous. The concerned technical paper is based on results of a module out of complete research work, where T5-base model is fine-tuned to generate summary of the medical text reports. This paper presents a study based on comparison of abstractive methods using fine-tuned T5-base model, Gemini 2.0 Flash language model and different extractive approaches. It also compares results achieved by the fine-tuned T5-base model with Gemini model of Google. Through the results, the authors claim that the T5-base model that is fine-tuned on medical reports’ dataset is the most suitable and effective choice for the said task amongst several existing options.