Identifying gross post-mortem organ images using a pre-trained convolutional neural network.

Area of Science:

• Digital pathology
• Artificial intelligence in medicine
• Forensic imaging analysis

Background:

• Convolutional neural networks (CNNs) are effective for organ and pathology identification in radiological and microscopic images.
• Limited research exists on applying CNNs to post-mortem gross images of visceral organs.

Purpose of the Study:

• To evaluate the feasibility of using a pre-trained CNN for classifying post-mortem gross images of visceral organs.
• To demonstrate high classification accuracy with small post-mortem image datasets.

Main Methods:

• Utilized 537 gross post-mortem images of brain, heart, lung, liver, spleen, and kidney.
• Employed a pre-trained CNN (Xception) with randomly divided training and testing datasets.
• Trained and tested the CNN on the prepared image datasets.

Main Results:

• Achieved overall accuracies greater than 95% for both training and testing datasets.
• Attained an F1 score greater than 0.95 for all dissected organs.
• Demonstrated successful classification of post-mortem images even with limited data.

Conclusions:

• Pre-trained CNNs can classify small datasets of post-mortem images with very high accuracy.
• This approach holds significant potential for data mining, education, research, and quality assurance in forensic pathology.
• Highlights a novel application of AI in the analysis of post-mortem gross pathology.