Medical Image Classification via Quality Analysis and Anomaly Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current microscopic pathological image analysis relies heavily on manual review, leading to low efficiency, high costs, and low annotation reliability, with a high likelihood of introducing invalid or unqualified data into the diagnostic process.
Innovation Solution
A medical image classification method using a pre-trained deep learning network for anomaly detection and classification, which performs quality analysis on medical images to extract feature information and filter out unqualified images based on attributes like hue, saturation, lightness, definition, and entropy, thereby improving data quality and reducing noise.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If manual review is used for medical image analysis, then annotation reliability is maintained through human expertise, but processing efficiency is low and costs are high
Solution Approach 1:
A deep learning network is introduced as an intermediary between manual annotation and final diagnosis. The network performs automated quality analysis and anomaly detection on medical images, filtering out low-quality images before they reach human annotators. This intermediary system handles the bulk of preliminary processing, allowing human experts to focus only on cases requiring their specialized judgment, thereby improving both efficiency and maintaining reliability.
Solution Approach 2:
The system performs preliminary quality analysis and anomaly detection using automated algorithms before human review. By pre-filtering images based on quality metrics (sharpness, contrast, noise levels) and detecting obvious anomalies through computer vision, the system prepares the data in advance, reducing the workload for human annotators and ensuring that only high-quality images requiring expert assessment are manually reviewed.
2Reliability
If manual annotation is performed on all medical images, then annotation reliability is maintained, but the cost and time consumption increase significantly
Solution Approach 1:
The system extracts and removes low-quality images from the dataset through automated quality analysis. By identifying images that fail to meet quality thresholds (based on metrics such as blur, noise, and artifact detection), the system extracts only the high-quality subset that requires manual annotation. This extraction process eliminates the need for time-consuming manual review of obviously poor-quality images while maintaining reliability on the images that are annotated.
3Productivity
If deep learning network is used for automated classification, then processing efficiency is improved, but the risk of introducing invalid data increases
Solution Approach 1:
The system implements a feedback mechanism where the deep learning network's automated classification results are evaluated against quality metrics and anomaly detection outcomes. Images that are classified but fail quality thresholds or show anomalies are flagged and either reprocessed or sent for manual review. This feedback loop ensures that automated processing efficiency does not compromise data quality, as problematic cases are identified and handled appropriately.
Data Source
AI summary
This application describes a medical image classification method, a model training method, and a server. The medical image classification method includes: obtaining, by a device, a medical image data set. The device includes a memory storing instructions and a processor in communication with the memory. The method includes performing, by the device, quality analysis on the medical image data set, to extract feature information of a medical image in the medical image data set; and classifying, by the device, the medical image data set based on the feature information and by using a pre-trained deep learning network for performing anomaly detection and classification, to obtain a classification result.