JoVE - Journal of Visualized Experiments
JoVE Visualize
Contact Us

Fully automated 3D machine learning model for HPV status characterization in oropharyngeal squamous cell carcinomas based on CT images

  • 0Boston University Chobanian & Avedisian School of Medicine, Boston, MA, United States of America.
American Journal of Otolaryngology +

|

May 4, 2024

|

May 4, 2024

View abstract on PubMed

Summary

This summary is machine-generated.

A deep learning model accurately predicts human papillomavirus (HPV) status in oropharyngeal cancer using CT scans. This automated approach shows promise for non-invasive clinical management of HPV-related oropharyngeal squamous cell carcinoma (OPSCC).

Area Of Science

  • Artificial Intelligence in Oncology
  • Medical Imaging Analysis
  • Head and Neck Cancer Research

Background

  • Human papillomavirus (HPV) status is a critical prognostic factor for oropharyngeal squamous cell carcinoma (OPSCC).
  • Accurate HPV status determination is essential for guiding treatment decisions and predicting patient survival.
  • Current methods for HPV status assessment can be invasive or time-consuming.

Purpose Of The Study

  • To evaluate the performance of a fully automated 3D convolutional neural network (CNN) for predicting HPV status in OPSCC patients.
  • To assess the accuracy of the CNN model in characterizing HPV status directly from pretreatment CT images.
  • To explore the potential of AI-driven imaging analysis as a non-invasive tool for HPV status determination.

Main Methods

  • A 3D DenseNet-121 model was trained using pretreatment CT images from OPSCC patients.
  • The model was developed to predict HPV-p16 status, a surrogate marker for HPV infection.
  • Model performance was rigorously evaluated using metrics including AUC, sensitivity, specificity, PPV, NPV, and F1 score.

Main Results

  • The 3D CNN achieved a mean Area Under the ROC Curve (AUC) of 0.80 ± 0.06.
  • The best-performing model iteration demonstrated high specificity (0.92) and sensitivity (0.86) at the Youden's index.
  • Excellent Positive Predictive Value (PPV) of 0.97 and F1 score of 0.82 were observed, alongside an NPV of 0.71.

Conclusions

  • A fully automated 3D CNN can accurately predict HPV status in OPSCC patients with high sensitivity and specificity.
  • This AI-based approach offers a promising non-invasive method for HPV status characterization.
  • Further development of this algorithm could significantly aid in the clinical management of OPSCC.

Keywords: