基于MRI圖像神經(jīng)網(wǎng)絡分析方法的前列腺癌診斷
發(fā)布時間:2023-06-23 19:00
前列腺癌是毀滅性的惡性腫瘤,早期很難識別。診斷的“金標準”是MRI和MRI掃描的進一步研究。人工神經(jīng)網(wǎng)絡具有巨大的圖像識別任務潛力,可用于自動診斷系統(tǒng),對醫(yī)療人員有很大幫助。在這項工作中,基于健康和不健康前列腺的MRI圖像,開發(fā)了用于識別前列腺癌的卷積神經(jīng)網(wǎng)絡模型。卷積多層單向神經(jīng)網(wǎng)絡的建議結構包括兩個卷積層和兩個池化層的交替,然后是三個完全連接的層。作為激活功能,除了輸出層之外,所有層都使用了Re LU功能。對于輸出層,使用了Soft Max激活功能。損失函數(shù)由MSE函數(shù)表示。選擇SGD函數(shù)作為優(yōu)化函數(shù)。收集數(shù)據(jù)并為訓練神經(jīng)網(wǎng)絡作好初步準備。用于訓練神經(jīng)網(wǎng)絡的數(shù)據(jù)集包括5450個樣本。在具有不同癌癥和健康樣本的三個不同數(shù)據(jù)集上測試了性能,并獲得了良好的結果。實驗表明,訓練集的準確率為90.5%–94.3%,測試集的準確率為89.2–96.9%。測試準確率和損失的曲線表明該模型已被很好地訓練。在某些情況下,準確率可能達到97.1%。具有一定的臨床應用價值,這種深度學習方法可以廣泛應用于前列腺癌及其他癌癥任務的分級和分期。該研究將使前列腺癌的診斷過程自動化,提高確定癌性腫瘤的準確性,減輕...
【文章頁數(shù)】:63 頁
【學位級別】:碩士
【文章目錄】:
摘要
Abstract
Nomenclature
Chapter 1 Introduction
1.1 Statement of the problem
1.2 Research background
1.3 Application of machine learning methods in medicine
1.4 Related works review
1.5 Content organization
Chapter 2 Theories Of Machine Learning
2.1 Deep learning and neural networks
2.2 Overview of popular image recognition methods
2.2.1 Histogram of Oriented Gradient
2.2.2 Recurrent neural network
2.3 Using convolution neural network for image recognition
2.3.1 Convolution
2.3.2 Pooling
2.3.3 Fully connected layers and training process
2.4 Development environment and frameworks
2.5 Chapter Summary
Chapter 3 Research Method And Structure Of CNN
3.1 Architecture
3.2 Optimization
3.2.1 Stochastic Gradient Descent (SGD)
3.2.2 RMSProp
3.2.3 ADAM
3.3 Loss function
3.4 Activation function
3.5 Dataset
3.6 Chapter summary
Chapter 4 Research experiment
4.1 Data preprocessing
4.2 Experiment conditions
4.3 Results and evaluation
Conclusion
結論
References
Acknowledgements
本文編號:3835140
【文章頁數(shù)】:63 頁
【學位級別】:碩士
【文章目錄】:
摘要
Abstract
Nomenclature
Chapter 1 Introduction
1.1 Statement of the problem
1.2 Research background
1.3 Application of machine learning methods in medicine
1.4 Related works review
1.5 Content organization
Chapter 2 Theories Of Machine Learning
2.1 Deep learning and neural networks
2.2 Overview of popular image recognition methods
2.2.1 Histogram of Oriented Gradient
2.2.2 Recurrent neural network
2.3 Using convolution neural network for image recognition
2.3.1 Convolution
2.3.2 Pooling
2.3.3 Fully connected layers and training process
2.4 Development environment and frameworks
2.5 Chapter Summary
Chapter 3 Research Method And Structure Of CNN
3.1 Architecture
3.2 Optimization
3.2.1 Stochastic Gradient Descent (SGD)
3.2.2 RMSProp
3.2.3 ADAM
3.3 Loss function
3.4 Activation function
3.5 Dataset
3.6 Chapter summary
Chapter 4 Research experiment
4.1 Data preprocessing
4.2 Experiment conditions
4.3 Results and evaluation
Conclusion
結論
References
Acknowledgements
本文編號:3835140
本文鏈接:http://sikaile.net/kejilunwen/zidonghuakongzhilunwen/3835140.html
最近更新
教材專著