[Paper Review] Localization and Classification of Parasitic Eggs in Microscopic Images Using an EfficientDet Detector

Nouar AlDahoul, Hezerul Abdul Karim+2 more|arXiv (Cornell University)|Aug 3, 2022

Digital Imaging for Blood Diseases4 citations

TL;DR

This paper proposes a multi-modal EfficientDet-based detector with an EfficientNet-v2 backbone to localize and classify parasitic eggs in microscopic images across 11 categories. Trained on the novel Chula-ParasiteEgg-11 dataset of 11,000 images, the model achieves 92% accuracy and 93% F1 score, demonstrating high localization and classification performance for automated parasitic egg detection in low-resource settings.

ABSTRACT

IPIs caused by protozoan and helminth parasites are among the most common infections in humans in LMICs. They are regarded as a severe public health concern, as they cause a wide array of potentially detrimental health conditions. Researchers have been developing pattern recognition techniques for the automatic identification of parasite eggs in microscopic images. Existing solutions still need improvements to reduce diagnostic errors and generate fast, efficient, and accurate results. Our paper addresses this and proposes a multi-modal learning detector to localize parasitic eggs and categorize them into 11 categories. The experiments were conducted on the novel Chula-ParasiteEgg-11 dataset that was used to train both EfficientDet model with EfficientNet-v2 backbone and EfficientNet-B7+SVM. The dataset has 11,000 microscopic training images from 11 categories. Our results show robust performance with an accuracy of 92%, and an F1 score of 93%. Additionally, the IOU distribution illustrates the high localization capability of the detector.

Motivation & Objective

• To address diagnostic errors and inefficiencies in manual parasitic egg detection in low- and middle-income countries (LMICs).
• To develop an automated, fast, and accurate system for localizing and classifying parasitic eggs in microscopic images.
• To introduce a novel multi-modal learning detector using EfficientDet and EfficientNet-v2 for improved performance in parasitic egg recognition.
• To evaluate the model on a newly curated dataset, Chula-ParasiteEgg-11, containing 11,000 images across 11 parasite egg categories.

Proposed method

• The proposed method employs an EfficientDet detector with an EfficientNet-v2 backbone for feature extraction and object detection.
• Multi-scale feature fusion is applied across feature pyramid networks (FPN) to enhance detection of small parasitic eggs.
• The model is trained end-to-end on the Chula-ParasiteEgg-11 dataset using a multi-label classification head for 11 parasite egg categories.
• A secondary baseline model using EfficientNet-B7 with SVM is also evaluated for comparative performance.
• Data augmentation techniques are applied to improve generalization and robustness across diverse microscopic image conditions.
• Intersection over Union (IoU) is used to evaluate localization accuracy, with distribution analysis showing high precision in bounding box predictions.

Experimental results

Research questions

• RQ1 Can a deep learning-based detector achieve high accuracy and robust localization in identifying parasitic eggs across diverse microscopic images?
• RQ2 How does the performance of EfficientDet with EfficientNet-v2 compare to alternative models like EfficientNet-B7+SVM in parasitic egg classification?
• RQ3 To what extent does the proposed multi-modal detector generalize across the 11 distinct parasitic egg categories in the Chula-ParasiteEgg-11 dataset?
• RQ4 What is the localization capability of the detector, as measured by IoU distribution across predicted bounding boxes?

Key findings

• The proposed EfficientDet model with EfficientNet-v2 backbone achieved a classification accuracy of 92% on the Chula-ParasiteEgg-11 dataset.
• The model demonstrated an F1 score of 93%, indicating strong balance between precision and recall across all 11 parasite egg categories.
• The IoU distribution analysis confirmed high localization accuracy, with most predictions falling within high IoU thresholds.
• The model outperformed the EfficientNet-B7+SVM baseline in both accuracy and F1 score, validating the effectiveness of the detector architecture.
• The results indicate that the proposed method is suitable for deployment in low-resource diagnostic settings due to its speed, accuracy, and robustness.