Medical Image Analysis

Neurocognitive Disorders > Dementia > Ai-based brain volumetry and morphometry

Longitudinal automated brain volumetry vs. expert visual assessment of atrophy progression on MRI is robust but caution is advised

Max Gebest, Christel Weiß, Chang-Gyu Cho, Lucrezia Hausner, Lutz Frölich, Alex Förster, Nandhini Santhanam, Johann Fontana, Christoph Groden, Holger Wenz,*, Máté E. Maros,*,# 


Automated tools have been proposed to quantify brain volume for suspected dementia diagnoses. However, their robustness in longitudinal, real-life cohorts remains unexplored. We investigated if expert visual assessment (EVA) of atrophy progression is reflected by automated volumetric analyses (AVA) on sequential MR-imaging. 

We analyzed a random subset of 20 patients with two consecutive 3D T1-weighted examinations (median follow-up 4.0 years, LQ-UQ: 2.1-5.2, range: 0.2-10). Thirteen (65%) with cognitive decline, the remaining with other neuropsychiatric diseases. EVA was performed by two blinded neuroradiologists using a 3 or 5-point Likert scale for atrophy progression (scores 0-2: no, probable and certain progression or decrease, respectively) in dementia-relevant brain regions (frontal-, parietal-, temporal lobes, hippocampi, ventricles). 

Differences of AVA-volumes were normalized to baseline (delta). Inter-rater agreement of EVA scores was excellent (κ=0.92). AVA-delta and EVA showed significant global associations for the right hippocampus (p=0.035), left temporal lobe (p=0.0092), ventricle volume (p=0.0091) and a weak association for the parietal lobe (p=0.067). Post hoc testing revealed a significant link for the left hippocampus (p=0.039). 

In conclusion, the associations between volumetric deltas and EVA of atrophy progression were robust for certain brain regions. However, AVA-deltas showed unexpected variance, and therefore should be used with caution in individual cases, especially when acquisition protocols vary. 

Real-life evaluation and stress test  of  

Neurovascular > Stroke > Vasospasm Detection

ASNR 2018, O-283; Real-time decision support for cerebral vasospasm detection on conventional angiograms using deep learning

M Maros1, A Förster1, M Alzghloul1, E Neumaier-Probst1, C Cho1, J Böhme1, C Groden1, H Wenz1 

1University Medical Center Mannheim, Medical Faculty Mannheim of University Heidelberg, Germany, Mannheim, Baden-Württemberg

Purpose: Cerebral vasospasm (VP) is a life threatening condition with highly increased risk of death or permanent disability. VP detection on conventional angiographic (DSA) images requires multiple years of training and expert level knowledge. Here, we investigated the applicability of deep learning based image classifier for VP detection. 

Materials and methods: A retrospective cohort of 91 patients (57F; 34M) mean age 57 years (range: 23-86 yrs) undergoing cerebral DSA was retrieved from local database. Angiograms were reassessed by two independent blinded neuroradiologists. The single most representative ap series of internal carotid artery (left/right or both) were selected (n=140) and categorized as VP positive (n+=50) or negative (n-=90). The angiograms were randomly divided into training (112; 80%) and test sets (28; 20%). We re-trained and custom modified a deep convolutional neural network (CNN) pre-trained on ImageNet[1] with rectified linear units (ReLu) and a fully connected dense block using softmax activation[2]. ... 

Neurooncology > Brain Tumors > Glioblastoma & Metastases

DGNR 2018, Deep learning-based imaging classifier for improving differential diagnosis between primary and metastatic brain tumors

Maros ME1, Cho CGy1, Förster A1, Böhme J1, Alzghloul M1, Neumaier-Probst E1, von Deimling A3, Ratliff M2, Seiz-Rosenhagen M2, Hänggi D2, GrodenC1, WenzH1 

1Dept. of Neuroradiology, 2Clin. for Neurosurgery, University Medical Center Mannheim, Medical Faculty Mannheim of, 3Dept. Neuropathology, University Heidelberg, 1,2Mannheim/3Heidelberg, Baden-Württemberg, Germany

Purpose: The solely radiological differentiation between glioblastoma multiforme (GBM) and metastases (META) at the initial diagnosis can be extremely challenging, due to similar imaging morphology and scarce clinical data regarding primary tumor. Here, we investigated the applicability of deep learning-based image classifiers to improve this distinction. 

Materials and methods: We retrieved a retrospective cohort of 430 patients including 212 GBMs (119M/93F, mean age 63.8 years, range: 2-92 yrs) and 218 METAs (122M/96F, mean age 61.8 years, range: 23-86 yrs) undergoing cranial magnetic resonance imaging (cMRI) using institutional tumor protocol. ...