Optimized Transfer Learning With Hybrid Feature Extraction for Uterine Tissue Classification Using Histopathological...

Area of Science:

• Oncology
• Medical Imaging
• Computational Biology

Background:

• Endometrial cancer, or uterine cancer, significantly impacts female reproductive health.
• Histopathological image analysis is crucial for diagnosis, but challenges remain in modeling complex image relationships and handling cell appearance variations.
• Early detection of endometrial cancer is often hindered by limitations in current diagnostic methods.

Purpose of the Study:

• To develop a novel classification technique for uterine tissue analysis.
• To improve the accuracy and reliability of endometrial cancer diagnosis through advanced computational methods.
• To address limitations in modeling complex histopathological image features and cell appearance variations.

Main Methods:

• A novel transfer learning convolution neural network with artificial bald eagle optimization (TL-CNN with ABEO) was developed.
• Image preprocessing utilized a median filter, followed by enhancement with the multiple identities representation network (MIRNet) adapted by pelican crow search optimization (PCSO).
• Tissue segmentation was aided by segmentation quality assessment (SQA), and parameter selection by deep convolutional neural network (DCNN) trained with fractional PCSO (FPCSO). Feature extraction and cell classification were performed by TL-CNN trained with ABEO, integrating the bald eagle search (BES) and artificial hummingbird algorithm (AHA).

Main Results:

• The proposed ABEO + TL-CNN model achieved high diagnostic performance on the cancer image archive dataset.
• Achieved an accuracy of 89.59%, sensitivity of 90.25%, and specificity of 89.89%.
• Demonstrated effectiveness in handling cell appearance variations and complex relationships within histopathological images.

Conclusions:

• The developed TL-CNN with ABEO technique offers a promising advancement for accurate endometrial cancer classification.
• This method shows potential for improving early detection and diagnosis of uterine cancer.
• The integration of optimization algorithms like ABEO significantly enhances the performance of deep learning models in medical image analysis.