Optimizing Deep Learning Models for Ovarian Cancer Subtype Classification: A Systematic Evaluation of Architectures and Data Augmentation Strategies

Research Square (Research Square), Journal Year: 2025, Volume and Issue: unknown

Published: May 7, 2025

Ovarian cancer is a leading cause of cancer-related mortality among women, and accurate classification its subtypes critical for effective treatment planning. This study systematically investigates the impact different network architectures data augmentation strategies on ovarian subtype classification. We evaluate two baseline models (VGG ViT) propose an efficient hybrid model that integrates convolutional self-attention mechanisms to balance local feature extraction global context modeling. Furthermore, we conduct comprehensive assessment various techniques, including geometric, color, spatial transformations, determine their effects generalization. Additionally, compare pre-trained non-pre-trained analyze benefits transfer learning in this domain. To enhance interpretability, utilize Grad-CAM visualizations examine decision-making processes models. Our findings reveal while ViT exhibits superior generalization capabilities with pre-training, VGG remains competitive even without pre-training due strong inductive biases. Among tested strategies, geometric transformations significantly improve performance, whereas color-based augmentations show limited or degrade performance. The proposed achieves comparable accuracy maintaining smaller parameter scale faster training efficiency. In conclusion, provides key insights into selection techniques pathological image design framework offers interpretable approach classification, potential applications broader medical imaging tasks.

Language: Английский

Segmentation of ovarian cyst in ultrasound images using AdaResU-net with optimization algorithm and deep learning model

Scientific Reports, Journal Year: 2024, Volume and Issue: 14(1)

Published: Aug. 14, 2024

Ovarian cysts pose significant health risks including torsion, infertility, and cancer, necessitating rapid accurate diagnosis. Ultrasonography is commonly employed for screening, yet its effectiveness hindered by challenges like weak contrast, speckle noise, hazy boundaries in images. This study proposes an adaptive deep learning-based segmentation technique using a database of ovarian ultrasound cyst A Guided Trilateral Filter (GTF) applied noise reduction pre-processing. Segmentation utilizes Adaptive Convolutional Neural Network (AdaResU-net) precise size identification benign/malignant classification, optimized via the Wild Horse Optimization (WHO) algorithm. Objective functions Dice Loss Coefficient Weighted Cross-Entropy are to enhance accuracy. Classification types performed Pyramidal Dilated (PDC) network. The method achieves accuracy 98.87%, surpassing existing techniques, thereby promising improved diagnostic patient care outcomes.

Language: Английский