In this work, we present a computer-aided diagnosis system that uses deep learning and decision fusion to classify patients into one of three classes: “Likely Prostate Cancer," “Equivocal" and “Likely not Prostate Cancer." We impose the group “Equivocal" to reduce misclassifications by allowing for uncertainty, akin to prostate imaging reporting systems used by radiologists. We trained 3D convolutional neural networks to perform two binary patient-level classification tasks: classification of patients with/without prostate cancer and classification of patients with/without clinically significant prostate cancer. Networks were trained separately using volumetric T2-weighted images and apparent diffusion coefficient maps for both tasks. The probabilistic outputs of the resulting four trained networks were combined using majority voting followed by the max operator to classify patients into one of the three classes mentioned above. All networks were trained using patient-level labels only, which is a key advantage of our system since voxel-level tumour annotation is often unavailable due to the time and effort required of a radiologist. Our system was evaluated by retrospective analysis on a previously collected trial dataset. At a higher sensitivity setting, our system achieved 0.97 sensitivity and 0.31 specificity compared to an experienced radiologist who achieved 0.99 sensitivity and 0.12 specificity. At a lower sensitivity setting, our system achieved 0.78 sensitivity and 0.77 specificity compared to 0.76 sensitivity and 0.77 specificity for the experienced radiologist. We envision our system acting as a second reader in pre-biopsy screening applications.
Motivation: Focal therapy is an emerging low-morbidity treatment option for low-intermediate risk prostate cancer; however, challenges remain in accurately delivering treatment to specified targets and determining treatment success. Registered multi-parametric magnetic resonance imaging (MPMRI) acquired before and after treatment can support focal therapy evaluation and optimization; however, contouring variability, when defining the prostate, the clinical target volume (CTV) and the ablation region in images, reduces the precision of quantitative image-based focal therapy evaluation metrics. To inform the interpretation and clarify the limitations of such metrics, we investigated inter-observer contouring variability and its impact on four metrics.
Methods: Pre-therapy and 2-week-post-therapy standard-of-care MPMRI were acquired from 5 focal cryotherapy patients. Two clinicians independently contoured, on each slice, the prostate (pre- and post-treatment) and the dominant index lesion CTV (pre-treatment) in the T2-weighted MRI, and the ablated region (post-treatment) in the dynamic-contrast- enhanced MRI. For each combination of clinician contours, post-treatment images were registered to pre-treatment images using a 3D biomechanical-model-based registration of prostate surfaces, and four metrics were computed: the proportion of the target tissue region that was ablated and the target:ablated region volume ratio for each of two targets (the CTV and an expanded planning target volume). Variance components analysis was used to measure the contribution of each type of contour to the variance in the therapy evaluation metrics.
Conclusions: 14–23% of evaluation metric variance was attributable to contouring variability (including 6–12% from ablation region contouring); reducing this variability could improve the precision of focal therapy evaluation metrics.