Purpose
Since 2009 In a small town in Tuscany (Fig. 1), the Montignoso HEart and Lung Project (MHELP) offers to the general population a screening for the prevention of lung and cardiovascular diseases (CVD).
1826 subjects aged between 45 and 75 years, joined the program (52% of the target population), undergoes a series of clinical tests (Fig. 2) including a low-dose chest CT to assess lung and CVD risks.
To date only 86 deaths and 67 cardiac events have occurred in the enrolled population, These numbers...
Methods and materials
We created a model capable of estimating the aging of the heart by extracting interpretable deep features developing a multitasking model composed by 3D convolutional autoencoder (CAE) and a Artificial Neural Network (ANN) classifier (Fig. 3).
The role played by the CAE is to ensure that the deep latent features used for the age prediction can be represented on the input CT image.
ANN instead has the role of ensuring an appropriate combination of the aforementioned features for the age regression task.
To train the...
Results
The model loss function reach the minimum value after 2000 training epochs. The cooperative/competitive approach allows both an accurate estimation of the age (MAE=3y) (Fig.5, Fig. 6) and the extraction of deep features able to regenerate the cardiac CT images of each age class.
Figure 7 shows a case were the biological age prediction is higher than the chronological one. Interestingly, there are imaging features in the CT volume, such as coronary calcifications, that show an evidence of atherosclerosis process.
Conclusion
In this study we show that the estimation of the age from chest CT scans is feasible using a deep learning strategy.
The estimated age could represent a predictive biomarker of cardiac events by analyzing outliers between the estimated age and the expected age and their correlation with cardiovascular risk factors.
As a computer science analogy, if a MHELP subject is like a class object the artificial intelligence is a method to extract hidden information such as the biological age of the heart (fig. 8).
Personal information and conflict of interest
D. della Latta; Massa/IT - nothing to disclose N. Martini; Massa, PLEASE SELECT AN OPTION BELOW/IT - nothing to disclose C. Fabbri; Massa/IT - nothing to disclose A. Ripoli; Pisa/IT - nothing to disclose S. Chiappino; Massa/IT - nothing to disclose F. Avogliero; Massa/IT - nothing to disclose M. Emdin; Pisa/IT - nothing to disclose D. Chiappino; Massa/IT - nothing to disclose A. Aimo; Pisa/IT - nothing to disclose
References
Pröve, P. L., Jopp-van Well, E., Stanczus, B., Morlock, M. M., Herrmann, J., Groth, M., ... & der Mauer, M. A. (2019). Automated segmentation of the knee for age assessment in 3D MR images using convolutional neural networks. International journal of legal medicine, 133(4), 1191-1205.
Smith, S. M., Vidaurre, D., Alfaro-Almagro, F., Nichols, T. E., & Miller, K. L. (2019). Estimation of brain age delta from brain imaging. NeuroImage.
Andrea Ripoli, Daniele Della Latta, Nicola Martini and Dante Chiappino. SSQ11-08 Radiophenomics: A Machine Learning Approach...