Abstract:Unsupervised anomaly detection is a popular approach for the analysis of neuroimaging data as it allows to identify a wide variety of anomalies from unlabelled data. It relies on building a subject-specific model of healthy appearance to which a subject's image can be compared to detect anomalies. In the literature, it is common for anomaly detection to rely on analysing the residual image between the subject's image and its pseudo-healthy reconstruction. This approach however has limitations partly due to the pseudo-healthy reconstructions being imperfect and to the lack of natural thresholding mechanism. Our proposed method, inspired by Z-scores, leverages the healthy population variability to overcome these limitations. Our experiments conducted on FDG PET scans from the ADNI database demonstrate the effectiveness of our approach in accurately identifying Alzheimer's disease related anomalies.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is how to use unsupervised anomaly detection techniques to accurately identify abnormalities related to Alzheimer's disease (AD) in brain FDG PET (fluorodeoxyglucose positron - emission tomography) images. Traditional unsupervised anomaly detection methods usually rely on analyzing the residual images between patient images and their pseudo - healthy reconstructions. This method has several limitations: 1. **Lack of a natural threshold mechanism**: This leads to a high false - positive rate. 2. **Failure to consider normal variations in the data set**: This makes it more difficult to detect subtle abnormalities. 3. **Failure to take into account the reconstruction error of the model itself**: Even in the images of healthy individuals, significant reconstruction errors may occur. To solve these problems, the authors propose a new method, which draws on the concept of Z - score and constructs an anomaly map by using the variability of the healthy population. Specifically, this method is implemented through the following steps: 1. **Generate pseudo - healthy images**: Use a variational auto - encoder (VAE) to generate pseudo - healthy images for each patient. 2. **Calculate the anomaly map**: By comparing the original image and the pseudo - healthy image and combining the standard deviation of the healthy images in the training set, generate an anomaly map. The formula for calculating the anomaly map is: \[ \tilde{z}=\frac{x - \hat{x}}{\sigma} \] where \(x\) is the input image, \(\hat{x}\) is the pseudo - healthy reconstructed image, and \(\sigma\) is the standard deviation of the healthy images in the training set. 3. **Set the threshold**: By setting the threshold to identify significant abnormal regions, thereby reducing the influence of model reconstruction errors. The experimental results show that the Z - score - based anomaly map can more effectively highlight the simulated hypometabolic regions, reduce the possibility of false positives, and improve the accuracy of anomaly detection. This method has been verified in the FDG PET scans of the ADNI database, showing its potential in the early detection of Alzheimer - related brain abnormalities.

Leveraging healthy population variability in deep learning unsupervised anomaly detection in brain FDG PET

Evaluation of pseudo-healthy image reconstruction for anomaly detection with deep generative models: Application to brain FDG PET

A Novel Cascade Machine Learning Pipeline for Alzheimer’s Disease Identification and Prediction

Towards frugal unsupervised detection of subtle abnormalities in medical imaging

Brainomaly: Unsupervised Neurologic Disease Detection Utilizing Unannotated T1-weighted Brain MR Images

Anomaly‐based Alzheimer's disease detection using entropy‐based probability Positron Emission Tomography images

Towards Universal Unsupervised Anomaly Detection in Medical Imaging

Leveraging the Mahalanobis Distance to enhance Unsupervised Brain MRI Anomaly Detection

Bias in Unsupervised Anomaly Detection in Brain MRI

Reversing the Abnormal: Pseudo-Healthy Generative Networks for Anomaly Detection

Unsupervised Anomaly Detection in 3D Brain MRI using Deep Learning with Multi-Task Brain Age Prediction

Simple statistical methods for unsupervised brain anomaly detection on MRI are competitive to deep learning methods

Patched Diffusion Models for Unsupervised Anomaly Detection in Brain MRI

Unsupervised anomaly detection in MR images using multicontrast information

Exploiting Epistemic Uncertainty of Anatomy Segmentation for Anomaly Detection in Retinal OCT

Feasibility of Universal Anomaly Detection without Knowing the Abnormality in Medical Images

TauAD: MRI-free Tau Anomaly Detection in PET Imaging via Conditioned Diffusion Models

Cross Attention Transformers for Multi-modal Unsupervised Whole-Body PET Anomaly Detection

GAN-based Multiple Adjacent Brain MRI Slice Reconstruction for Unsupervised Alzheimer's Disease Diagnosis

Unsupervised Anomaly Detection in 3D Brain MRI using Deep Learning with impured training data

Unsupervised Analysis of Alzheimer's Disease Signatures using 3D Deformable Autoencoders