HEp-2 image classification using a multi-class and multiple-binary classifier.

Area of Science:

• Medical diagnostics
• Artificial Intelligence
• Immunofluorescence

Background:

• Accurate identification of indirect immunofluorescence patterns in human epithelial type 2 (HEp-2) cells is crucial for diagnosing autoimmune diseases.
• Manual interpretation of HEp-2 cell images presents challenges including subjectivity, inconsistency, and inefficiency.

Purpose of the Study:

• To develop an automated method for classifying HEp-2 cell images to overcome the limitations of manual interpretation.
• To propose a novel Multi-Class and Multiple-Binary Classifier (MCMBC) for improved HEp-2 cell image classification.

Main Methods:

• The proposed MCMBC is an ensemble learner combining a multi-class (MC) sub-classifier using a multi-scale convolutional neural network (MSCNN) and multiple-binary (MB) sub-classifiers utilizing pre-trained VGG16 networks.
• The MB sub-classifiers specifically address easy-to-confuse class pairs identified by the MC sub-classifier.
• Final predictions integrate features from the MC sub-classifier and outputs from the MB sub-classifiers.

Main Results:

• The MCMBC model demonstrated superior performance compared to state-of-the-art methods on the ICPR 2014 Task-2 dataset.
• The model achieved higher average classification accuracy (ACA) (84.68% vs. 83.35%) and mean classification accuracy (MCA) (82.89% vs. 82.67%).

Conclusions:

• The developed MCMBC model offers a more accurate and efficient approach to HEp-2 cell image classification.
• This automated method holds significant potential for improving the diagnostic process of autoimmune diseases.