Detection of Emerging Infectious Diseases in Lung CT based on Spatial Anomaly Patterns

TL;DR

This work first identifies anomalies in lung CT data, and then compares their distribution in a continually acquired new patient cohorts with historic patient population observed over a long prior period to detect the emergence of new disease phenotypes exhibiting distinct patterns of the spatial distribution of lesions.

Abstract

Fast detection of emerging diseases is important for containing their spread and treating patients effectively. Local anomalies are relevant, but often novel diseases involve familiar disease patterns in new spatial distributions. Therefore, established local anomaly detection approaches may fail to identify them as new. Here, we present a novel approach to detect the emergence of new disease phenotypes exhibiting distinct patterns of the spatial distribution of lesions. We first identify anomalies in lung CT data, and then compare their distribution in a continually acquired new patient cohorts with historic patient population observed over a long prior period. We evaluate how accumulated evidence collected in the stream of patients is able to detect the onset of an emerging disease. In a gram-matrix based representation derived from the intermediate layers of a three-dimensional convolutional neural network, newly emerging clusters indicate emerging diseases.

Figures (4)

• Figure 1: To detect a novel disease phenotype, we compute anomaly maps and extract features using a pre-trained 3D U-Net ronneberger2015uzhou2021models. We then calculate the gram-matrix and multiply it with the intermediate feature representation to obtain a disease appearance descriptor. An autoencoder-based dimensionality reduction provides the final case embeddings. This distribution helps identify new clusters of cases that differ from previous patient populations while being similar to each other.
• Figure 2: (a) The introduction of previously unseen diseases is shown for two different rates in orange and red and compared to the steady flow of known diseases. (b) t-SNE visualization of the embedding of a patient population with 5 different diseases. (c) and (d) The average days $n_{ped}$ and $n_{kde}$ necessary for positive detection as a function of reproduction number $R$ for three window sizes $w_s$.
• Figure 3: Analogous to Fig. \ref{['fig: combofig0']}, the results for real data are in panels e) to g).
• Figure 4: Embedding spaces capturing the population with 5 simulated diseases illustrating the separation of different diseases: (h) intermediate layer output features; (i) gram-matrix; (j) proposed feature representation.