PI-CAI: Concurrent AI (Workshop)
The PI-CAI Challenge was an extensive confirmatory study providing substantial evidence that artificial intelligence (AI) exhibits superior standalone performance in diagnosing and detecting clinically significant prostate cancer (csPCa) compared to prostate radiologists in a retrospective setting.
The PI-CAI Concurrent AI Reader Study examines the added value of AI assistance during prostate MRI assessment through a multi-center, multinational reader study. The study encompasses two reading phases, separated by a washout period.
Instructions ✏️¶
This reading block contains 20 cases (+/-50% bpMRI only and +/-50% bpMRI with AI outcomes). The cases are presented in alternation between conventional readings and AI-supported readings. During AI assistance, the sixth image port contains an AI score for csPCa and a csPCa lesion detection map.
Besides bpMRI and AI outcomes, readers receive clinical (age, PSA, prostate volume and PSAD) and technical (vendor and high-b value) variables.
For each case perform the following actions:
▶️ Point annotate and score csPCa lesions (if present) using a PI-RADS 3-5 score
📣 A separate multiple point annotation question is defined for each PI-RADS grade (3 in total). Utilize the assigned question related to a specific PI-RADS score to annotate lesions corresponding to the designated grade.
▶️ Provide an overall patient-level suspicion score for the presence of csPCa, between 0–100 (with incremental steps of 1).
📣 Please note, this is not a confidence score but a likelihood score of csPCa presence
Important Remarks 🔊¶
◽️ Refer to the AI System Specifications in the instruction manual to get the most out of your AI-assisted readings
◽️ Readers cannot revisit any cases
◽️ Use of absolute ADC values to distinguish between benign and malignant prostate tissues is discouraged considering that ADC values can vary substantially depending on selected b-values, vendor, image scaling and inter-patient variability. (T. Barrett et al., 2015)
◽️ Reading time is expected to be 3-5 min per case.
◽️ Reading progress will be saved. Hence, readers can complete readings in multiple sessions.
◽️ Please be aware that your reading time will be noted as timestamps of your answers are recorded.