Automatic CIN Grades Prediction of Sequential Cervigram Image Using LSTM With Multistate CNN Features.

Area of Science:

• Oncology
• Medical Imaging
• Artificial Intelligence

Background:

• Cervical cancer is a leading global cancer in women.
• Colposcopy is crucial for screening cervical intraepithelial neoplasia (CIN) and cancer but suffers from high misdiagnosis rates.
• Current computer-assisted diagnostic tools overlook the sequential, multistate nature of colposcopy, limiting clinical utility.

Purpose of the Study:

• To develop an advanced AI model for accurate classification of CIN grades and cervical cancer using cervigram images.
• To address the limitations of existing algorithms by incorporating the temporal and multistate aspects of colposcopy.
• To improve the diagnostic accuracy of cervical cancer screening.

Main Methods:

• A novel cervigram-based recurrent convolutional neural network (C-RCNN) was constructed.
• Convolutional neural networks extracted spatial features from images.
• A sequence-encoding module captured temporal features, and a multistate-aware convolutional layer integrated features from different colposcopic states.

Main Results:

• The C-RCNN achieved 96.13% test accuracy on a dataset of 4,753 cervigrams.
• High specificity (98.22%) and sensitivity (95.09%) were obtained.
• Areas under the receiver operating characteristic curves exceeded 0.94, demonstrating effective joint optimization of visual and sequential data.

Conclusions:

• The C-RCNN effectively classifies cervical cancer and CIN grades by analyzing sequential cervigram image dynamics.
• This AI approach significantly outperforms methods that analyze single image frames.
• The C-RCNN architecture shows potential for broader applications in medical image analysis.