Modinat Aina Abayomi

Leon Hamnett, Mary Adewunmi, Modinat Abayomi et al.

Breast cancer remains a critical global health challenge, necessitating early and accurate detection for effective treatment. This paper introduces a methodology that combines automated image augmentation selection (RandAugment) with search optimisation strategies (Tree-based Parzen Estimator) to identify optimal values for the number of image augmentations and the magnitude of their associated augmentation parameters, leading to enhanced segmentation performance. We empirically validate our approach on breast cancer histology slides, focusing on the segmentation of cancer cells. A comparative analysis of state-of-the-art transformer-based segmentation models is conducted, including SegFormer, PoolFormer, and MaskFormer models, to establish a comprehensive baseline, before applying the augmentation methodology. Our results show that the proposed methodology leads to segmentation models that are more resilient to variations in histology slides whilst maintaining high levels of segmentation performance, and show improved segmentation of the tumour class when compared to previous research. Our best result after applying the augmentations is a Dice Score of 84.08 and an IoU score of 72.54 when segmenting the tumour class. The primary contribution of this paper is the development of a methodology that enhances segmentation performance while ensuring model robustness to data variances. This has significant implications for medical practitioners, enabling the development of more effective machine learning models for clinical applications to identify breast cancer cells from histology slides. Furthermore, the codebase accompanying this research will be released upon publication. This will facilitate further research and application development based on our methodology, thereby amplifying its impact.

Reem Abdel-Salam, Mary Adewunmi, Modinat A. Abayomi

Large language models (LLMs) are increasingly evident for accurate question answering across various domains. However, rigorous evaluation of their performance on complex question-answering (QA) capabilities is essential before deployment in real-world biomedical and healthcare applications. This paper presents our approach to the MedHopQA track of the BioCreative IX shared task, which focuses on multi-hop biomedical question answering involving diseases, genes, and chemicals. We adopt a supervised fine-tuning strategy leveraging LLaMA 3 8B, enhanced with a curated biomedical question-answer dataset compiled from external sources including BioASQ, MedQuAD, and TREC. Three experimental setups are explored: fine-tuning on combined short and long answers, short answers only, and long answers only. While our models demonstrate strong domain understanding, achieving concept-level accuracy scores of up to 0.8, their Exact Match (EM) scores remain significantly lower, particularly in the test phase. We introduce a two-stage inference pipeline for precise short-answer extraction to mitigate verbosity and improve alignment with evaluation metrics. Despite partial improvements, challenges persist in generating strictly formatted outputs. Our findings highlight the gap between semantic understanding and exact answer evaluation in biomedical LLM applications, motivating further research in output control and post-processing strategies.