Utilizing a publicly accessible automated machine learning platform to enable diagnosis before tumor surgery

Area of Science:

• Radiology
• Oncology
• Artificial Intelligence

Background:

• Sinonasal inverted papilloma (IP) can transform into squamous cell carcinoma (IP-SCC), necessitating different surgical approaches.
• Pre-operative biopsy sampling errors can complicate diagnosis and surgical planning for IP.
• Artificial intelligence (AI) offers a potential solution for differentiating IP from IP-SCC.

Purpose of the Study:

• To develop and evaluate an AI model for distinguishing between IP and IP-SCC using CT imaging.
• To assess the accuracy of AI in identifying malignant transformation pre-operatively.

Main Methods:

• A deep learning image classification model was trained on CT images from 958 patients across 19 institutions.
• The Google Cloud Vertex AI platform was utilized for model training and validation.
• The model's performance was evaluated on a holdout test dataset using metrics like AUC, sensitivity, specificity, accuracy, and F1 score.

Main Results:

• The AI model was trained and validated on 41,099 individual CT images.
• The model achieved 95.8% sensitivity and 99.7% specificity in identifying IP-SCC.
• An overall accuracy of 99.1% was obtained for the deep learning model.

Conclusions:

• A deep automated machine learning model can accurately identify malignant transformation of IP using pre-operative CT imaging.
• The AI model, developed using a publicly available tool, shows excellent accuracy in differentiating IP from IP-SCC.
• This AI approach can aid in pre-operative diagnosis and surgical planning for sinonasal tumors.