Predicting Lymph Node Metastasis From Primary Cervical Squamous Cell Carcinoma Based on Deep Learning in...

Area of Science:

• Oncology
• Medical Imaging
• Artificial Intelligence

Background:

• Cervical cancer lymph node metastasis is a critical prognostic factor.
• Accurate preoperative assessment of lymph node status is vital for treatment planning.
• Current methods for lymph node staging can be invasive and may not always be precise.

Purpose of the Study:

• To develop and validate a deep learning framework for predicting lymph node metastasis in cervical cancer.
• To assess the framework's performance using both internal and external datasets.
• To explore the potential of AI in improving preoperative staging of cervical cancer.

Main Methods:

• A multi-instance deep convolutional neural network with a multiscale attention mechanism was developed.
• The framework was trained on 1524 hematoxylin and eosin-stained whole slide images (WSIs) from 564 cervical cancer patients.
• Performance was evaluated on independent internal (Fudan University Shanghai Cancer Center) and external (Cancer Genome Atlas) test sets.

Main Results:

• The framework achieved an area under the receiver operating characteristic curve (AUC) of 0.87 in cross-validation for predicting lymph node metastasis.
• AUCs of 0.84 (internal test set) and 0.75 (external test set) demonstrated good predictive performance.
• A retrained network accurately predicted para-aortic lymph node metastasis in patients with positive pelvic nodes (AUCs of 0.91 and 0.88).

Conclusions:

• Deep learning analysis of primary tumor pathological images shows potential for preoperative assessment of cervical cancer lymph node status.
• The developed framework offers a non-invasive approach to aid in surgical and treatment decisions.
• Further validation with cervical biopsy specimens and larger multicenter datasets is necessary to confirm clinical utility.