DL May Confuse Calcifications with ICH Heterogeneity, Warns Letter

A letter in World Journal of Radiology warns deep learning models for spontaneous ICH may mistake basal ganglia calcifications for hematoma heterogeneity and chronic lacunar infarcts for perihematomal edema, inflating false positives. Radiologist-verified quantitative radiomics…

• Allowing CNNs to detect density 50–400 HU without restriction may falsely classify age-related and pathological basal ganglia calcifications as heterogeneity.
• Not restricting to 25–35 HU for perihematomal edema detection may lead to mistaking microischemic areas and chronic lacunar infarcts for PHE.
• Hand-crafted quantitative radiomics by trained radiologists can measure ICH and PHE volume more accurately than CNNs using only three slices.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.