Deep Learning Artefact Detection for Histology Slides
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods fail to efficiently detect and reject histology slides with artefacts early in the process flow, leading to time-consuming and resource-intensive rework and potential misdiagnoses.
Innovation Solution
Implement a preliminary imaging and analysis step using a deep learning model to identify artefacts before scanning, allowing for automatic detection and rejection of slides with rejectable artefacts, thereby ensuring only artefact-free slides proceed to detailed analysis.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If detailed image analysis is performed on all histology slides, then diagnostic accuracy is improved, but time consumption and resource usage increase significantly
Solution Approach 1:
The patent implements a preliminary quality assessment step using a deep learning model to detect artefacts in histology slides before they undergo detailed image analysis. This preliminary action filters out slides with rejectable artefacts, ensuring that only high-quality slides proceed to the time-consuming detailed analysis stage, thus reducing overall time consumption while maintaining diagnostic accuracy for valid slides
Solution Approach 2:
The patent extracts and removes slides containing rejectable artefacts from the analysis pipeline using automated deep learning-based detection. By separating problematic slides from the main workflow before detailed analysis, the system prevents waste of time and resources on unusable samples while preserving the integrity of the diagnostic process for quality slides
2Measurement precision
If manual artefact detection is performed by pathologists, then diagnostic precision is maintained, but productivity decreases due to additional work
Solution Approach 1:
The patent implements a self-service quality control system where a deep learning model automatically performs artefact detection and classification on histology slides. The system autonomously identifies rejectable artefacts and flags slides for removal from the analysis pipeline, eliminating the need for pathologists to manually inspect each slide for artefacts while preserving their role in diagnosing quality slides
Solution Approach 2:
The patent introduces an intermediary deep learning-based quality assessment system between slide preparation and pathologist review. This intermediary layer pre-screens slides for artefacts, filtering out problematic samples before they reach pathologists, thus protecting diagnostic precision while significantly increasing pathologist throughput by reducing their workload
3Reliability
If slides with artefacts are processed through the full workflow, then no diagnostic errors occur, but resource waste increases
Solution Approach 1:
The patent performs preliminary artefact detection using deep learning models before slides enter the resource-intensive processing workflow. By identifying and removing slides with rejectable artefacts at this early stage, the system prevents unnecessary consumption of computational resources, storage space, and processing time on unusable slides while maintaining diagnostic reliability for valid samples
Solution Approach 2:
The patent implements an automated mechanism to discard slides containing rejectable artefacts based on deep learning-based quality assessment. The system identifies problematic slides, removes them from further processing, and prevents waste of resources on these unusable samples, while preserving resources for high-quality slides that can contribute to accurate diagnostics
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The present disclosure relates to a method of analyzing a plurality of histology slides, wherein possible artefacts are detected at an early stage. This is achieved by including a preliminary imaging and image analysis step in the process flow; the step being executed preferably before any diagnostic assessment e.g. image analysis or manual reading is performed by a pathologist or a lab technician. Accordingly, the histology slides reaching the expert for image analysis are of a higher quality using the presently disclosed method, since the slides are ideally free of artefacts. The method disclosed herein thus saves valuable time for the pathologist, since only artefact-free slides will be subject to a detailed analysis. Furthermore, the method according to the present disclosure minimizes the risk of misinterpretations leading to potentially false diagnoses. The present disclosure further relates to a deep learning model capable of automatically determining whether histopathological images are suitable for diagnostic and/or research assessment, and it further relates to the training of said deep learning model.