Classification of suspicious lesions on prostate multiparametric MRI using machine learning

Abstract

We present a radiomics-based approach developed for the SPIE-AAPM-NCI PROSTATEx challenge. The task was to classify clinically significant prostate cancer in multiparametric (mp) MRI. Data consisted of a "training dataset" (330 suspected lesions from 204 patients) and a "test dataset" (208 lesions/140 patients). All studies included T2-weighted (T2-W), proton density-weighted, dynamic contrast enhanced, and diffusionweighted imaging. Analysis of the images was performed using the MIM imaging platform (MIM Software, Cleveland, Ohio). Prostate and peripheral zone contours were manually outlined on the T2-W images. A workflow for rigid fusion of the aforementioned images to T2-W was created in MIM. The suspicious lesion was outlined using the high b-value image. Intensity and texture features were extracted on four imaging modalities and characterized using nine histogram descriptors: 10%, 25%, 50%, 75%, 90%, mean, standard deviation, kurtosis, and skewness (216 features). Three classification methods were used: classification and regression trees (CART), random forests, and adaptive least absolute shrinkage and selection operator (LASSO). In the held out by the organizers test dataset, the areas under the curve (AUCs) were: 0.82 (random forests), 0.76 (CART), and 0.76 (adaptive LASSO). AUC of 0.82 was the fourth-highest score of 71 entries (32 teams) and the highest for feature-based methods.

Radiomic and Genomic Machine Learning Method Performance for Prostate Cancer Diagnosis: Systematic Literature Review

“…Figure 3 shows the subgroup analysis among studies that employed internal cross-validation techniques (subgroup 1) [ 46 - 49 , 51 , 54 , 55 , 59 ], split validation approaches (subgroup 2) [ 44 , 50 , 52 , 53 , 56 , 57 ], and no validation (subgroup 3) [ 45 , 58 ]. The heterogeneity for subgroups 1 and 2 was around 80% ( P <.001).…”

“…Figure 4 shows the subgroup analysis for regression-based models (subgroup 1) [ 45 , 48 , 49 , 51 , 56 - 58 ], tree-based models (subgroup 2) [ 46 , 53 , 59 ], and DL methods (subgroup 3) [ 44 , 47 , 50 , 52 , 55 ]. One study was not included [ 54 ], as it was the only study employing a support vector machine model.…”

Semantic learning machine improves the CNN-Based detection of prostate cancer in non-contrast-enhanced MRI

“…The test data from the PROSTATEx competition were not usable for our purposes since the ground-truth labels (i.e., CS or non-CS) are not provided [2]. It is worth noting that the PROSTATEx challenge uses biopsy points as a ground-truth instead of the corresponding tumor segmentations [20,47]. By taking into account the available MRI sequences in the PROSTATEx dataset as well as by following the current clinical trend that aims at reducing the use of contrast medium administration [49], in this study we analyzed only the non-contrast-enhanced MRI sequences, namely: T2w, PDw, and ADC series.…”

“…This joint approach was compared against 2D U-Net [35] and 3D DeepMedic [22], obtaining the highest classification accuracy. However, the automated analysis of radiomics features, based on traditional machine learning models [24], is gaining interest in the assessment of the heterogeneity in PCa [46].…”

Biopsy-free prediction of prostate cancer aggressiveness using deep learning and radiology imaging

“…In another related study, Kwon et al described a radiomics-based approach to classify clinically significant lesions in multi-parametric MRI (mp-MRI) using three feature-based methods: regression trees, random forests, and a regularization techniques for simultaneous estimation and variable selection (adaptive LASSO). Random forest achieved highest performance with an AUC of 0.82 24 . Recently Rubinstein et al used an unsupervised deep learning method to detect and localize prostate tumors in PET/CT images 25 .…”