Low-Cost Histopathological Mitosis Detection for Microscope-acquired Images.

Area of Science:

• Pathology
• Medical Imaging
• Artificial Intelligence

Background:

• Cancer outcomes are poor in resource-limited countries due to high costs and limited pathologists.
• Digital pathology aids cancer diagnosis, but Whole Slide Image scanners are unaffordable in low-income regions.
• Microscope-acquired images offer a cost-effective alternative for automated cancer detection.

Purpose of the Study:

• To propose LCH-Network, a novel method for identifying cancer mitotic counts from microscope-acquired images.
• To address data imbalance and varying image scales for improved mitotic localization.

Main Methods:

• Developed LCH-Network, incorporating Label Mix and Generative Adversarial Networks (GANs) for image synthesis.
• Applied progressive resolution to effectively handle different image scales.
• Utilized microscope-acquired images for mitotic count estimation.

Main Results:

• Achieved an F1-Score of 0.71, outperforming existing techniques.
• Demonstrated the efficacy of LCH-Network in mitotic localization.
• Validated the feasibility of low-cost mitotic count estimation.

Conclusions:

• LCH-Network provides a viable solution for mitotic count estimation using affordable microscopic images.
• This method can enhance cancer diagnosis accuracy in resource-limited settings.
• Clinical application can prevent presumptive treatment without confirmed diagnosis.