Lightweight Truncated Fused-MirrorNet for Classification and Analysis of Histopathology Images

Area of Science:

• Computational pathology
• Artificial intelligence in oncology
• Medical image analysis

Background:

• Renal cell carcinoma (RCC) is a leading cause of cancer mortality, underscoring the need for precise early diagnosis.
• Manual histopathology image classification is labor-intensive, time-consuming, and prone to inter-observer variability, risking misdiagnosis.
• Existing deep learning models often require substantial computational resources, limiting their accessibility in low-resource settings.

Purpose of the Study:

• To develop an efficient and accurate deep learning approach for automated kidney histopathology image classification.
• To create a lightweight model deployable on low-end infrastructure, addressing computational constraints.
• To enhance the performance and reduce the training time compared to current state-of-the-art models.

Main Methods:

• Implementation of a lightweight, truncated Fused-MirrorNet model utilizing a mirrored architecture.
• Application of partial layer freezing and feature fusion techniques to optimize model performance.
• Training and evaluation on two distinct histopathology image datasets: TCGA kidney and BreakHis.

Main Results:

• The proposed Fused-MirrorNet model demonstrated superior performance over existing Convolutional Neural Network (CNN) and vision transformer models.
• Achieved high classification accuracy: 92.60% on the TCGA kidney dataset and 90.00% on the BreakHis dataset.
• Significantly reduced training time while maintaining high classification accuracy, indicating improved efficiency.

Conclusions:

• The Fused-MirrorNet model offers a deployable, scalable, and reproducible solution for kidney histopathology image analysis.
• The developed approach simplifies the creation of vision-based deep learning models, removing the need for complex computational methods.
• This lightweight model effectively addresses the limitations of manual classification and resource-intensive AI, facilitating broader adoption in cancer diagnostics.