A machine learning‐based quality control tool for functional MRI images

Background Changes in functional MRI (fMRI) have been associated with Alzheimer’s disease and other neurological disorders. Quality control (QC) of images prior to their computational analysis is a key step clinical trial workflow, it traditionally implemented visually by image analysts. However, the number acquired from large studies makes neuroimaging QC difficult task that diverts skilled resources introduces inter-rater variability. This problem accentuated fMRI, where analysts need inspect 4D sequences, although processing reports can be used as aid, having each report remains. Method We an automated fMRI process using stacking classifier (STC), random forest + support vector machine, imaging quality metrics (IQM) inputs. The IQMs were estimated MRIQC. also added designed IQM measures how centred participant’s brain within image. As train/test dataset we ABIDE multisite database. For our experiments, only adult brains database used, leaving 14 sites available for leave-one-site-out (LOSO) cross-validation experiment. All included experiment QC’d validated protocols at IXICO. Result LOSO validation procedure resulted mean pass-vs-fail accuracy 92% (with 5.7% standard deviation, SD). Pass/fail sensitivity specificity 97% (6.7% SD) 69% (29% respectively. tested fully trained STC model on independent healthy ageing participants OpenNeuro (ds002872), which 87% accuracy, 94% 50% specificity, validates modelling procedure. Conclusion It possible design deploy automatic fMRI. Although was high all models average low, highlighting difficulty detecting fails pipeline. deployed detect clear passes will not visual QC, reliably reducing cases require inspection. Our future work focus testing developed additional datasets (e.g. ADNI) validate its performance before roll-out applications. aging after age 90.