Authors: Neel Kanwal, Farbod Khoraminia, Umay Kiraz, Andres Mosquera-Zamudio, Carlos Monteagudo, Emiel A.M. Janssen, Tahlita C.M. Zuiverloon, Chunmig Rong, Kjersti Engan

Published on: March 12, 2024

Impact Score: 7.4

Arxiv code: Arxiv:2403.07743

Summary

• What is new: The research introduces a novel application of a mixture of experts (MoE) scheme for identifying artifacts in whole slide images (WSIs) for cancer diagnosis, focusing on five types of artifacts.
• Why this is important: Histopathology for cancer diagnosis faces challenges due to artifacts in WSIs, which can lead to incorrect predictions by deep learning algorithms.
• What the research proposes: A mixture of experts (MoE) scheme is proposed, using independent binary deep learning models to detect specific artifacts and a fusion mechanism to ensemble predictions.
• Results: The MoE schemes, particularly the DCNNs-based MoE, outperform simpler multiclass models, achieving 86.15% F1 and 97.93% sensitivity scores on unseen data.

Technical Details

Technological frameworks used: Deep Convolutional Neural Networks (DCNNs), Vision Transformers (ViTs)

Models used: Mixture of experts (MoE), independent binary deep learning models

Data used: WSIs from different hospitals and cancer types

Potential Impact

Healthcare diagnostics, digital pathology companies, and AI diagnostics software developers

Want to implement this idea in a business?

We have generated a startup concept here: ClearPath AI.