Prediction and Mapping of Intraprostatic Tumor Extent with Artificial Intelligence

Abstract

Background: Magnetic resonance imaging (MRI) underestimation of prostate cancer extent complicates the definition of focal treatment margins.

Objective: To validate focal treatment margins produced by an artificial intelligence (AI) model.

Design setting and participants: Testing was conducted retrospectively in an independent dataset of 50 consecutive patients who had radical prostatectomy for intermediate-risk cancer. An AI deep learning model incorporated multimodal imaging and biopsy data to produce three-dimensional cancer estimation maps and margins. AI margins were compared with conventional MRI regions of interest (ROIs), 10-mm margins around ROIs, and hemigland margins. The AI model also furnished predictions of negative surgical margin probability, which were assessed for accuracy.

Outcome measurements and statistical analysis: Comparing AI with conventional margins, sensitivity was evaluated using Wilcoxon signed-rank tests and negative margin rates using chi-square tests. Predicted versus observed negative margin probability was assessed using linear regression. Clinically significant prostate cancer (International Society of Urological Pathology grade ≥2) delineated on whole-mount histopathology served as ground truth.

Results and limitations: The mean sensitivity for cancer-bearing voxels was higher for AI margins (97%) than for conventional ROIs (37%, p < 0.001), 10-mm ROI margins (93%, p = 0.24), and hemigland margins (94%, p < 0.001). For index lesions, AI margins were more often negative (90%) than conventional ROIs (0%, p < 0.001), 10-mm ROI margins (82%, p = 0.24), and hemigland margins (66%, p = 0.004). Predicted and observed negative margin probabilities were strongly correlated (R² = 0.98, median error = 4%). Limitations include a validation dataset derived from a single institution's prostatectomy population.

Conclusions: The AI model was accurate and effective in an independent test set. This approach could improve and standardize treatment margin definition, potentially reducing cancer recurrence rates. Furthermore, an accurate assessment of negative margin probability could facilitate informed decision-making for patients and physicians.

Patient summary: Artificial intelligence was used to predict the extent of tumors in surgically removed prostate specimens. It predicted tumor margins more accurately than conventional methods.

Keywords: Artificial intelligence; Magnetic resonance imaging; Prostatic neoplasms; Surgical margins; Surgical pathology.

Fig. 1

Overview of artificial intelligence (AI) model input and output, including: (A) input data, which consists of T2-weighted magnetic resonance imaging (MRI), serum prostate-specific antigen (PSA), and biopsy core locations with pathology labels; (B) high-level model architecture, wherein image features are generated via a convolutional neural network and other features are engineered from biopsy and pathology data; and (C) application of the model to produce a cancer estimation map (CEM) showing voxel-level predictions of clinically significance prostate cancer (csPCa). The CEM is thresholded in order to produce a treatment margin.

Fig. 2

Plot of encapsulation confidence score (ECS) versus margin volume (% of the prostate) and selection of default margins for two exemplary cases. (A) A small tumor and “cool” cancer estimation map (CEM). (B) A large tumor and “hot” CEM. Cancer estimation maps are shown on the left, with the default margin outlined in pink. Corresponding whole-mount pathology images are shown in the middle, with clinically significant tumor regions highlighted in black. (C) ECS versus margin volume curves and default margin selection for the same exemplary cases.

Fig. 3

Two exemplary cases from the independent test dataset where artificial intelligence (AI) margins were negative, having succeeded in encapsulating extensions of the index lesion that were invisible on magnetic resonance imaging (MRI): (A–C) a case for which hemigland margins were positive and (D–F) a case for which 10-mm MRI region of interest (ROI) margins were positive. Figures 3A and 3D show input data including MRI, biopsy, and projected ROI location; Figures 3B and 3E show the cancer estimation map and default AI margin; and Figures 3C and 3F show whole mount histopathology along with segmentation of clinically significant prostate cancer (csPCa).

Fig. 4

Calibration curve plotting predicted versus observed negative margin rates alongside a linear regression fit (R² = 0.98). The 95% confidence bands cover a deviation of ±8% in negative margin rates.