AI-Assisted Detection of Challenging Ischemic Stroke on DWI: A Reader Study

A dataset of 250 selected MRI diffusion-weighted imaging (DWI) cases (130 challenging acute ischemic stroke, 120 controls) used to evaluate AI-assisted diagnostic performance. The data originates from a retrospective, single-center study of 3,986 patients conducted between February 2017 and November 2021. The dataset was authored by Younbeom Jeong and shared under a CC-BY-4.0 license.

Use Cases

• Benchmarking AI model performance for stroke lesion segmentation based on Dice similarity coefficient (DSC) metrics mentioned in the description
• Studying the impact of AI assistance on radiologist diagnostic confidence based on the reported improvement in challenging cases
• Analyzing diagnostic performance metrics (AUC, sensitivity, specificity) for stroke detection from the multi-reader study results
• Investigating false-negative stroke cases identified by AI, as referenced in the description

Strengths

• Includes performance metrics from a multi-reader study with five readers, providing comparative human vs. AI data
• Reports specific quantitative results, including AI sensitivity of 96.0% and improvement in lesion segmentation DSC from 0.523 to 0.742
• Based on a substantial initial cohort of 3,986 patients, from which the study cases were selected

Limitations

• Column-level documentation is absent; field semantics must be inferred after download
• Row count for the core 250-case dataset is unknown, which may limit suitability assessment
• Data may reflect temporal and single-center bias inherent to the source study

Provenance

Source

figshare, author Younbeom Jeong

Collection Method

Retrospective, single-center, randomized crossover study

Time Range

February 2017 to November 2021

Freshness

Last updated 2026-05-21 04:31:24; freshness should be verified