A Hybrid Intelligence Approach for Circulating Tumor Cell Enumeration in Digital Pathology by Using CNN and Weak...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computational Biology

Background:

• Manual counting of Circulating Tumor Cells (CTCs) for cancer screening is labor-intensive.
• Deep learning (DL) shows promise in medical image analysis.

Purpose of the Study:

• To develop a hybrid AI system for automated CTC enumeration.
• To combine cytopathologist expertise with DL Convolutional Neural Networks (CNNs).

Main Methods:

• Developed a three-component hybrid system: CNN-based detection/localization (weak annotations), CNN-based segmentation, and Support Vector Machine (SVM) classification.
• Utilized B-scale transform for sphericality, incorporating size, texture, and area for SVM input.
• Trained and validated on fluoroscopic images for CTC detection, segmentation, and enumeration.

Main Results:

• Achieved high precision (0.98) and recall (0.92) for CTC detection, comparable to state-of-the-art methods.
• Demonstrated a low counting error of 2-3% (with B-scale) on an independent dataset.
• Outperformed previous thresholding methods (30% error) significantly.

Conclusions:

• The hybrid AI approach offers an efficient and accurate method for CTC enumeration.
• This system has the potential to reduce the time and cost associated with cancer screening.
• Further research in object localization and segmentation is supported by these findings.