No abstract

Deep neural networks are increasingly used for neurological disease classification by MRI, but the networks' decisions are not easily interpretable by humans. Heat mapping by deep Taylor decomposition revealed that (potentially misleading) image features even outside of the brain tissue are crucial for the classifier's decision. We propose a regularization technique to train convolutional neural network (CNN) classifiers utilizing relevance-guided heat maps calculated online during training. The method was applied using T1-weighted MR images from 128 subjects with Alzheimer's disease (mean age=71.9+-8.5 years) and 290 control subjects (mean age=71.3+-6.4 years). The developed relevance-guided framework achieves higher classification accuracies than conventional CNNs but more importantly, it relies on less but more relevant and physiological plausible voxels within brain tissue. Additionally, preprocessing effects from skull stripping and registration are mitigated, rendering this practically useful in deep learning neuroimaging studies. Understanding the decision mechanisms underlying CNNs, these results challenge the notion that unprocessed T1-weighted brain MR images in standard CNNs yield higher classification accuracy in Alzheimer's disease than solely atrophy.

Background: While iron is essential for normal brain functioning, elevated concentrations are commonly found in neurodegenerative diseases and are associated with impaired cognition and neurological deficits. Currently, only little is known about genetic and environmental factors that influence brain iron concentrations. Methods: Heritability and bivariate heritability of regional brain iron concentrations, assessed by R2* relaxometry at 3 Tesla MRI, were estimated with variance components models in 130 middle-aged to elderly participants of the Austrian Stroke Prevention Family Study. Results: Heritability of R2* iron ranged from 0.46 to 0.82 in basal ganglia and from 0.65 to 0.76 in cortical lobes. Age and BMI explained up to 12% and 9% of the variance of R2* iron, while APOE ε4 carrier status, hypertension, diabetes, hypercholesterolemia, sex and smoking explained 5% or less. The genetic correlation of R2* iron among basal ganglionic nuclei and among cortical lobes ranged from 0.78 to 0.87 and from 0.65 to 0.97, respectively. R2* rates in basal ganglia and cortex were not genetically correlated. Conclusions: Regional brain iron concentrations are mainly driven by genetic factors while environmental factors contribute to a certain extent. Brain iron levels in the basal ganglia and cortex are controlled by distinct sets of genes.

Deep neural networks are increasingly used for neurological disease classification by MRI, but the networks’ decisions are not easily interpretable by humans. Heat mapping by deep Taylor decomposition revealed that (potentially misleading) image features even outside of the brain tissue are crucial for the classifier’s decision. We propose a regularization technique to train convolutional neural network (CNN) classifiers utilizing relevance-guided heat maps calculated online during training. The method was applied using T1-weighted MR images from 128 subjects with Alzheimer’s disease (mean age = 71.9 ± 8.5 years) and 290 control subjects (mean age = 71.3 ± 6.4 years). The developed relevance-guided framework achieves higher classification accuracies than conventional CNNs but more importantly, it relies on less but more relevant and physiological plausible voxels within brain tissue. Additionally, preprocessing effects from skull stripping and registration are mitigated. With the interpretability of the decision mechanisms underlying CNNs, these results challenge the notion that unprocessed T1-weighted brain MR images in standard CNNs yield higher classification accuracy in Alzheimer’s disease than solely atrophy.