Automatic identification of clinically important Aspergillus species by artificial intelligence-based image...

Area of Science:

• Clinical Mycology
• Medical Diagnostics
• Artificial Intelligence in Medicine

Background:

• Traditional Aspergillus identification relies on morphological examination, requiring mycological expertise.
• Advanced methods like PCR-sequencing and MALDI-TOF MS are accurate but require specialized personnel and expensive equipment.
• Artificial intelligence (AI), particularly image recognition, is emerging as a tool for rapid, automated medical diagnoses.

Purpose of the Study:

• To explore the potential utility of AI-based image recognition for identifying Aspergillus species.
• To evaluate the performance and accuracy of different convolutional neural networks (CNNs) for automated identification using colonial images.

Main Methods:

• A proof-of-concept study utilized a dataset of 6867 colonial images from four clinically significant Aspergillus species.
• Images were divided into training (2813), validation (2814), and testing (1240) sets.
• Three CNNs—ResNet-18, Inception-v3, and DenseNet-121—were trained and evaluated for their identification accuracy.

Main Results:

• ResNet-18 demonstrated the highest testing accuracy (99.35%) with the fewest misidentifications (n=8), outperforming Inception-v3 and DenseNet-121.
• The accuracy of identification was positively correlated with the distinctiveness of morphological features in the images.
• AI-based image recognition proved effective for Aspergillus identification, with unique morphological features enhancing accuracy.

Conclusions:

• AI-based image recognition using colonial images is a promising technology for Aspergillus identification.
• This approach offers a rapid, user-friendly, and potentially cost-effective alternative to conventional methods.
• With further database expansion, AI holds potential as a routine diagnostic tool in clinical laboratories.