Deep Convolution Neural Network for Malignancy Detection and Classification in Microscopic Uterine Cervix Cell Images

Area of Science:

• Medical Imaging
• Computational Pathology
• Artificial Intelligence in Healthcare

Background:

• Automated Pap smear screening is crucial for cervical cancer detection.
• Traditional methods struggle with large, diverse datasets due to limitations in hand-engineered features.
• Image preprocessing and segmentation pose challenges in pathology.

Purpose of the Study:

• To develop a deep learning model for accurate classification of cervical cell images.
• To overcome limitations of traditional methods in Pap smear analysis.
• To enhance automated cancer detection using Convolution Neural Networks (CNNs).

Main Methods:

• A novel hierarchical architecture using Convolution Neural Networks (CNNs) for feature extraction.
• The CNN model learns visual features like edges, size, shape, and colors.
• A deep prediction model was built and trained on an augmented Herlev dataset for classifying five grades of cervical cancer.

Main Results:

• The model achieved high accuracy across different class problems: 94.1% for 5 classes, 96.2% for 4 classes, 94.8% for 3 classes, and 95.7% for 2 classes using enhanced original images.
• Contour extracted images yielded accuracies up to 94.7% for 3 classes.
• Binary images achieved the highest accuracy of 99.97% for 2 classes.

Conclusions:

• The proposed deep learning model demonstrates effective classification of cervical cell images with varying cancer grades.
• The study highlights the significant potential of deep learning in advancing Pap smear analysis and cancer detection.
• The findings support the use of CNNs for robust and accurate automated cervical cancer screening.