Comments on Eusebio et al.: Voxel-based analysis of whole-brain effects of age and gender on dopamine transporter SPECT imaging in healthy subjects

Data-driven identification of diagnostically useful extrastriatal signal in dopamine transporter SPECT using explainable AI

This study used explainable artificial intelligence for data-driven identification of extrastriatal brain regions that can contribute to the interpretation of dopamine transporter SPECT with ¹²³I-FP-CIT in parkinsonian syndromes. A total of 1306 ¹²³I-FP-CIT-SPECT were included retrospectively. Binary classification as ‘reduced’ or ‘normal’ striatal ¹²³I-FP-CIT uptake by an experienced reader served as standard-of-truth. A custom-made 3-dimensional convolutional neural network (CNN) was trained for classification of the SPECT images with 1006 randomly selected images in three different settings: “full image”, “striatum only” (3-dimensional region covering the striata cropped from the full image), “without striatum” (full image with striatal region removed). The remaining 300 SPECT images were used to test the CNN classification performance. Layer-wise relevance propagation (LRP) was used for voxelwise quantification of the relevance for the CNN-based classification in this test set. Overall accuracy of CNN-based classification was 97.0%, 95.7%, and 69.3% in the “full image”, “striatum only”, and “without striatum” setting. Prominent contributions in the LRP-based relevance maps beyond the striatal signal were detected in insula, amygdala, ventromedial prefrontal cortex, thalamus, anterior temporal cortex, superior frontal lobe, and pons, suggesting that ¹²³I-FP-CIT uptake in these brain regions provides clinically useful information for the differentiation of neurodegenerative and non-neurodegenerative parkinsonian syndromes.

Explainable AI to improve acceptance of convolutional neural networks for automatic classification of dopamine transporter SPECT in the diagnosis of clinically uncertain parkinsonian syndromes

Purpose

Deep convolutional neural networks (CNN) provide high accuracy for automatic classification of dopamine transporter (DAT) SPECT images. However, CNN are inherently black-box in nature lacking any kind of explanation for their decisions. This limits their acceptance for clinical use. This study tested layer-wise relevance propagation (LRP) to explain CNN-based classification of DAT-SPECT in patients with clinically uncertain parkinsonian syndromes.

Methods

The study retrospectively included 1296 clinical DAT-SPECT with visual binary interpretation as “normal” or “reduced” by two experienced readers as standard-of-truth. A custom-made CNN was trained with 1008 randomly selected DAT-SPECT. The remaining 288 DAT-SPECT were used to assess classification performance of the CNN and to test LRP for explanation of the CNN-based classification.

Results

Overall accuracy, sensitivity, and specificity of the CNN were 95.8%, 92.8%, and 98.7%, respectively. LRP provided relevance maps that were easy to interpret in each individual DAT-SPECT. In particular, the putamen in the hemisphere most affected by nigrostriatal degeneration was the most relevant brain region for CNN-based classification in all reduced DAT-SPECT. Some misclassified DAT-SPECT showed an “inconsistent” relevance map more typical for the true class label.

Conclusion

LRP is useful to provide explanation of CNN-based decisions in individual DAT-SPECT and, therefore, can be recommended to support CNN-based classification of DAT-SPECT in clinical routine. Total computation time of 3 s is compatible with busy clinical workflow. The utility of “inconsistent” relevance maps to identify misclassified cases requires further investigation.

Supplementary Information

The online version contains supplementary material available at 10.1007/s00259-021-05569-9.