Authors: Abdullah Alsalemi, Anza Shakeel, Mollie Clark, Syed Ali Khurram, Shan E Ahmed Raza

Published on: May 03, 2024

Impact Score: 7.4

Arxiv code: Arxiv:2405.01937

Summary

• What is new: The use of a vision transformer based Mask R-CNN for lesion detection and a Multiple Instance Learning scheme for classification in the context of head and neck cancer detection.
• Why this is important: Delayed diagnosis of head and neck cancer due to missed detections in surgical biopsies.
• What the research proposes: An attention-based pipeline using advanced machine learning models for detecting, segmenting, and classifying lesions as non-dysplastic, dysplastic, or cancerous.
• Results: Achieved up to 82% overlap accuracy in segmentation on unseen external test data and an F1-score of 85% for classification on the internal cohort test set.

Technical Details

Technological frameworks used: Mask R-CNN with a vision transformer approach for segmentation, and a Multiple Instance Learning based scheme for classification.

Models used: Vision transformer based Mask R-CNN, Multiple Instance Learning (MIL).

Data used: Clinical images for segmentation and classification, including a future approach to include endoscopic video data.

Potential Impact

Healthcare providers and diagnostic centers, Medical imaging software companies, Smart device makers with health monitoring capabilities.

Want to implement this idea in a business?

We have generated a startup concept here: EarlyDetectAI.