Related Experiment Videos
Improving Renal Cell Carcinoma Classification by Automatic Region of Interest Selection
Qaiser Chaudry1, S Hussain Raza2, Yachna Sharma3
1Georgia Institute of Technology, Atlanta, GA 30332 USA (phone: 404-542-2998; qaiser@gatech.edu ).
Proceedings. IEEE International Symposium on Bioinformatics and Bioengineering
|April 11, 2017
Summary
This study enhances automated renal cell carcinoma classification by developing a region of interest selection method. This approach improves accuracy by excluding non-diagnostic tissue, aiding pathologists in cancer diagnosis.
Area of Science:
- Pathology
- Medical Imaging
- Computer-Aided Diagnosis
Background:
- Manual analysis of renal cell carcinoma biopsies is challenging due to tissue variability and inter-observer errors.
- Existing automated systems struggle with necrotic regions and glands in clinical biopsy images.
- Previous work introduced a knowledge-based automated system for classification.
Purpose of the Study:
- To improve the accuracy of automated pathological image classification for renal cell carcinoma.
- To address the degradation in classification performance caused by necrotic regions and glands.
- To develop a region of interest (ROI) selection method mimicking pathologist techniques.
Main Methods:
- Proposed a novel, automated region of interest (ROI) selection process.
- The ROI selection automatically identifies and excludes necrotic regions and glands.
- Implemented this technique within an existing knowledge-based automated classification system.
Main Results:
- Achieved a significant improvement in classification accuracy from 90% to 95%.
- Demonstrated enhanced performance on a heterogeneous dataset of renal cell carcinoma images.
- The ROI selection effectively mitigated the negative impact of non-diagnostic tissue components.
Conclusions:
- The proposed ROI selection method substantially improves automated renal cell carcinoma classification accuracy.
- This technique offers a more robust and reliable automated solution for pathological image analysis.
- The findings support the integration of pathologist-inspired ROI selection in diagnostic AI systems.