Searching for prostate cancer by fully automated magnetic resonance imaging classification: deep learning versus non-deep learning

Abstract

Prostate cancer (PCa) is a major cause of death since ancient time documented in Egyptian Ptolemaic mummy imaging. PCa detection is critical to personalized medicine and varies considerably under an MRI scan. 172 patients with 2,602 morphologic images (axial 2D T2-weighted imaging) of the prostate were obtained. A deep learning with deep convolutional neural network (DCNN) and a non-deep learning with SIFT image feature and bag-of-word (BoW), a representative method for image recognition and analysis, were used to distinguish pathologically confirmed PCa patients from prostate benign conditions (BCs) patients with prostatitis or prostate benign hyperplasia (BPH). In fully automated detection of PCa patients, deep learning had a statistically higher area under the receiver operating characteristics curve (AUC) than non-deep learning (P = 0.0007 < 0.001). The AUCs were 0.84 (95% CI 0.78–0.89) for deep learning method and 0.70 (95% CI 0.63–0.77) for non-deep learning method, respectively. Our results suggest that deep learning with DCNN is superior to non-deep learning with SIFT image feature and BoW model for fully automated PCa patients differentiation from prostate BCs patients. Our deep learning method is extensible to image modalities such as MR imaging, CT and PET of other organs.

Prostate Cancer Detection using Deep Convolutional Neural Networks

“…Moreover, most of ROI-based methods use sliding windows of pixels as data structures to feed the CNNs, which makes it a challenging task to achieve an acceptable performance on classification of PCa at patient level due to the fact that each patient's MRI data constitutes several hundreds of thousands of windows of pixels. Therefore, ROI-based methods 22,24,26 struggle to merge individual ROI-based results into patient-level classification and they usually rely on basic merging methods such as simple voting 29 , which makes it a challenging task to achieve acceptable performance at patient level.…”

“…It is generally a challenging task to merge ROI-based or slice-level results into patient-level results 22-24, 26, 27 . Wang et al 29 compared the performance of deep learning-based methods to non-deep learning-based methods on the classification of PCa MRI slices vs non PCa MRI slices with 172 patients. They evaluated their VGGNet inspired 7 layers (5 convolution layers and 2 inner product layers) CNNs classifier's performance based on cross-validation.…”

Prostate cancer classification with multiparametric MRI transfer learning model

“…Most of them utilized conventional machine-learning techniques, which typically extract low-level radiomics features to characterize images and train a separate classifier. [4][5][6][7][8][9][10][11][12][13][14][15][16] Fehr et al 8 utilized first-order (mean, SD, skewness, and kurtosis) and second-order texture features (Haralick Features from gray level co-occurrence matrix (GLCM)) to assess the Gleason scores. Vignati et al 9 used the contrast and homogeneity of GLCM texture features on T2w images and ADC maps to help differentiate between patients with different Gleason scores.…”

“…Many methods have been proposed to classify prostate cancer. Most of them utilized conventional machine‐learning techniques, which typically extract low‐level radiomics features to characterize images and train a separate classifier . Fehr et al .…”

“…Many machine learning approaches [4][5][6][7][8][9][10][11][12][13][14][15][16] have been proposed for prostate cancer classification with MRI images. Most of these methods used low-level radiomics features based on the previous clinical experiences, which could not take full advantage of the complete information included in MRI images.…”