Convolutional Neural Networks for Automated Classification of Prostate Multiparametric Magnetic Resonance Imaging Based on Image Quality

Abstract

Background: Prostate magnetic resonance imaging (MRI) is technically demanding, requiring high image quality to reach its full diagnostic potential. An automated method to identify diagnostically inadequate images could help optimize image quality.

Purpose: To develop a convolutional neural networks (CNNs) based analysis pipeline for the classification of prostate MRI image quality.

Study type: Retrospective.

Subjects: Three hundred sixteen prostate mpMRI scans and 312 men (median age 67).

Field strength/sequence: A 3 T; fast spin echo T2WI, echo planar imaging DWI, ADC, gradient-echo dynamic contrast enhanced (DCE).

Assessment: MRI scans were reviewed by three genitourinary radiologists (V.P., M.D.M., S.C.) with 21, 12, and 5 years of experience, respectively. Sequences were labeled as high quality (Q1) or low quality (Q0) and used as the reference standard for all analyses.

Statistical tests: Sequences were split into training, validation, and testing sets (869, 250, and 120 sequences, respectively). Inter-reader agreement was assessed with the Fleiss kappa. Following preprocessing and data augmentation, 28 CNNs were trained on MRI slices for each sequence. Model performance was assessed on both a per-slice and a per-sequence basis. A pairwise t-test was performed to compare performances of the classifiers.

Results: The number of sequences labeled as Q0 or Q1 was 38 vs. 278 for T2WI, 43 vs. 273 for DWI, 41 vs. 275 for ADC, and 38 vs. 253 for DCE. Inter-reader agreement was almost perfect for T2WI and DCE and substantial for DWI and ADC. On the per-slice analysis, accuracy was 89.95% ± 0.02% for T2WI, 79.83% ± 0.04% for DWI, 76.64% ± 0.04% for ADC, 96.62% ± 0.01% for DCE. On the per-sequence analysis, accuracy was 100% ± 0.00% for T2WI, DWI, and DCE, and 92.31% ± 0.00% for ADC. The three best algorithms performed significantly better than the remaining ones on every sequence (P-value < 0.05).

Data conclusion: CNNs achieved high accuracy in classifying prostate MRI image quality on an individual-slice basis and almost perfect accuracy when classifying the entire sequences.

Evidence level: 4 TECHNICAL EFFICACY: Stage 1.

Keywords: artificial intelligence; deep learning; multiparametric MRI; prostate cancer; quality control.

FIGURE 1

Graphical representation of the analysis pipeline. Individual slices from a given sequence are preprocessed (including normalization and voxel resampling, and data augmentation) and subsequently fed to the CNN algorithm that assigns a classification label to every slice. Classification results for all slices from the same sequence are then aggregated by means of a majority vote aggregation function, so that a classification label is assigned to the entire acquired sequence.

FIGURE 2

Case examples of high‐ and low‐quality scans on T2WI images. It shows examples of high‐ and low‐quality T2 images: (a) high‐quality axial T2WI image (Q1), with good spatial resolution a tissue contrast; (b) low‐quality image (Q0) with poor spatial resolution and blurred details due to patient movement during acquisition, the sequence should be repeated in order to be able to accurately interpret the study; (c) very poor‐quality acquisition (Q0) due to evident magnetic susceptibility artifacts caused by a femoral prosthesis; (d) low‐quality image (Q0) because of inadequate S/N ratio making diagnostic accuracy suboptimal, the sequence needs to be repeated following optimization of the acquisition parameters.

FIGURE 3

Case examples of high‐ and low‐quality scans on DWI images. It shows examples of high‐ and low‐quality DWI images: (a) high‐quality DWI image (Q1), with good S/N ratio and no evident artifacts; (b) low‐quality image (Q0) with susceptibility artifacts caused by the presence of air in the rectum—the ability to detect foci in the right posterior peripheral zone is significantly impaired—the sequence could be repeated following attempts to expel the air from the rectum; (c) inadequate acquisition (Q0) with marked distortion and signal void due to magnetic susceptibility artifacts caused by a femoral prosthesis; (d) low‐quality image (Q0) because of inadequate S/N ratio that lower significantly the diagnostic power—the sequence needs to be repeated following optimization of the acquisition parameters.

FIGURE 4

Case examples of high‐ and low‐quality scans on DCE images. It shows examples of high‐ and low‐quality perfusion images: (a) high‐quality DCE image, with good contrast enhancement of the prostate gland; (b) low‐quality image (Q0) with low contrast enhancement of the prostate gland and high noise significantly impairing the sensitivity to detect suspicious foci; (c) low‐quality acquisition (Q0) due to both low contrast enhancement of the prostate gland and to low S/N ratio, the diagnostic sensitivity of this sequence is limited; (d) poor‐quality image (Q0) because of marked motion artifacts—the ability to correctly identify areas of pathologic enhancement is compromised.