Fast pancreatic cancer diagnosis using lightweight Mamba framework

#4487

Introduction: Pancreatic cancer is a deadly malignancy often diagnosed too late for effective treatment. Fast and precise identification of pathological regions in histopathology images is essential but limited by current slow methods. We present a framework to accelerate this process and enhance diagnostic support.

Aim(s): Our aim is to develop a framework for TCGA-PAAD images to rapidly detect and classify pancreatic cancer pathology regions. Specifically, we use clustering for quick localisation, followed by an optimised lightweight Mamba architecture for classification, integrating text feature embeddings to improve accuracy.

Materials and methods: We used TCGA-PAAD histopathology images. The method starts with clustering to identify pathology regions swiftly. These are processed with an efficient Mamba architecture for accurate classification. Text-based features were embedded to enrich model performance. Model evaluation metrics included accuracy and speed.

Conference:

Presenting Author:

Authors: Shengzhe Y, Yan K,

Keywords: Pancreatic Cancer, Lightweight Mamba Architecture, Pathology Region Classification and Recognition,