MRI Driven Diagnostic Management Pathway - Task 2

Value Proposition

Multiparametric magnetic resonance imaging (mpMRI) of the prostate is well established as a diagnostic tool for detection and localization of prostate cancer, allowing for targeted sampling of MRI-visible lesions that meet image criteria for clinically significant cancer. Known limitations of mpMRI include dependence on optimized imaging quality and experience of the interpreting radiologist, as well as false negativity in MR-invisible prostate cancers and false positivity in mimickers of cancer. Thus, AI-driven diagnostic tool for automated detection and classification of prostatic lesions on mpMRI is needed to address the following challenges: 1) variable quality of mpMRI and quality-dependent degradation of imaging, 2) high inter-observer variability of human readers in the interpretation of mpMRI, 3) under-utilization of imaging data not perceptible by the human eye that could enhance detection capability of mpMRI. Successful delivery of the MRI-directed pathway for men with elevated PSA suspected of having prostate cancer relies on maximization of diagnostic capability of mpMRI. 

Narrative(s)

70-year-old man with PSA 8.5 ng/mL, negative DRE, prior TRUS biopsy was negative. 

Workflow Description

The relevant images are obtained from the modality and sent to PACS and the AI engine based on anatomic landmarks. 

Accepted diagnostic quality mpMRI is analyzed by AI-driven detection and classification tools. Detected region of interest (ROI) is anatomically localized/segmented within the prostate using the reference PI-RADS anatomical diagram. PI-RADS criteria for scoring abnormalities detected on mpMRI are applied to the ROI on T2-weighted images, diffusion-weighted images and dynamic contrast enhanced images. Overall PI-RADS score, for each lesion with PI-RADS score >=3, is computed based on PI-RADS 2.1 rules. PI-RADS scoring criteria for T2, DWI, DCE images and rules for the Overall PI-RADS score are described in the PI-RADS v2.1 document. 

Output: ROI with anatomical label and PI-RADS score (1-5 reflecting the likelihood that a clinically significant cancer is present). Maximum four ROIs with the highest PI-RADS score will be recorded.

Actions: ROI and PI-RADS score will be delivered to the radiologist for approval. Radiologists will accept or reject the ROI for the purpose of reporting. If accepted, ROI will be reported and used for targeting during biopsy. 

Labels: Reference standard for ROI will be consensus interpretation by at least 2 prostate MRI qualified radiologists (as per local quantification rules). 

Reference standard for diagnostic accuracy of the radiologist: Prostatectomy specimen or systematic plus targeted biopsy (saturation biopsy) will be used to define radiologist performance for detection of CS cancer (GG >= 2). True positive is PI-RADS >=3 lesion with corresponding GG >= 2 tissue at that site. True negative is no annotated lesions and biopsy showing GGin that region. Special case is a negative MRI and no biopsy. In such a case, a follow-up systematic biopsy anytime after the MRI showing no GG >= 2 cancer is acceptable proof of a true negative. 

AI-driven detection and classification tools will have non-inferior performance to reported radiologists’ performance. For the AI-driven tool to be deemed non-inferior to the human reader, it must be within 5% of the human rader performance in detection of, -GG1 detection of more significant cancers and biopsy avoidance. 

An algorithm evaluated the MRI and based on anatomic landmarks for the location of the prostate, determines whether the likelihood that cancer is present, absent, or indeterminate by assessing DWI, ADC and DCE MRI. Based on the findings a PI-RADS score will be output:

• PI-RADS 1 - Very low (clinically significant cancer is highly unlikely to be present)

• PI-RADS 2 - Low (clinically significant cancer is unlikely to be present)

• PI-RADS 3 - Intermediate (the presence of clinically significant cancer is equivocal)

• PI-RADS 4 - High (clinically significant cancer is likely to be present)

• PI-RADS 5 - Very high (clinically significant cancer is highly likely to be present)

Inputs

DICOM Study

Primary Outputs

Detection of Diagnostic versus Non-Diagnostic