Related Experiment Video
Updated: Aug 5, 2026
Quantifying Microglia Morphology from Photomicrographs of Immunohistochemistry Prepared Tissue Using ImageJ
Published on: June 5, 2018
Systematic investigation of pre-processing and feature extraction techniques in medical image analysis
Pegah Dehbozorgi1,2, Oleg Ryabchykov1,2, Thomas W Bocklitz1,2
1Leibniz Institute of Photonic Technology, Member of Leibniz Health Technologies, Member of the Leibniz Centre for Photonics in Infection Research (LPI), Albert-Einstein-Strasse 9, 07745 Jena, Germany.
Optimizing medical image analysis requires tailored preprocessing and feature extraction. Combining these techniques significantly boosts diagnostic accuracy across radiology, pathology, and ophthalmology, with no single method universally superior.
Area of Science:
- Medical Imaging
- Computer-Aided Diagnosis
- Machine Learning
Background:
- Medical imagery is crucial for diagnosis, but its effectiveness relies on data quality and feature extraction.
- Optimizing image preprocessing and feature extraction is key to improving classification task performance.
Purpose of the Study:
- To evaluate the impact of various image preprocessing and feature extraction techniques on classification model performance.
- To assess these impacts across three distinct medical imaging modalities: radiology (chest X-rays), pathology (H&E patches), and ophthalmology (OCT scans).
Main Methods:
- Investigated nine preprocessing techniques (adjustment, filtering, normalization) combined systematically for each modality.
- Extracted features using five deep learning architectures (VGG16, ResNet50, DenseNet121, MobileNetV2, InceptionV3).
- Classified features using a Principal Component Analysis-Linear Discriminant Analysis (PCA-LDA) pipeline and evaluated using mean sensitivity.
Main Results:
- Significant improvements in mean sensitivity scores were observed: H&E images (74.9% to 96.95%), chest X-rays (89.9% to 96.65%), and OCT scans (82.4% to 98.8%).
- Optimal preprocessing configurations varied by modality, indicating no universal solution.
- VGG16 and DenseNet121 demonstrated the most robust performance across different settings.
Conclusions:
- Implementing at least one preprocessing step alongside a robust deep learning feature extractor substantially enhances model efficacy in diagnostic detection.
- Tailored preprocessing and feature extraction strategies are essential for optimizing medical image classification performance.
