(Predictable) performance bias in unsupervised anomaly detection.

BACKGROUND: With the ever-increasing amount of medical imaging data, the demand for algorithms to assist clinicians has amplified. Unsupervised anomaly detection (UAD) models promise to aid in the crucial first step of disease detection. While previous studies have thoroughly explored fairness in supervised models in healthcare, for UAD, this has so far been unexplored. METHODS: In this study, we evaluated how dataset composition regarding subgroups manifests in disparate performance of UAD models along multiple protected variables on three large-scale publicly available chest X-ray datasets. Our experiments were validated using two state-of-the-art UAD models for medical images. Finally, we introduced subgroup-AUROC (sAUROC), which aids in quantifying fairness in machine learning. FINDINGS: Our experiments revealed empirical "fairness laws" (similar to "scaling laws" for Transformers) for training-dataset composition: Linear relationships between anomaly detection performance within a subpopulation and its representation in the training data. Our study further revealed performance disparities, even in the case of balanced training data, and compound effects that exacerbate the drop in performance for subjects associated with multiple adversely affected groups. INTERPRETATION: Our study quantified the disparate performance of UAD models against certain demographic subgroups. Importantly, we showed that this unfairness cannot be mitigated by balanced representation alone. Instead, the representation of some subgroups seems harder to learn by UAD models than that of others. The empirical "fairness laws" discovered in our study make disparate performance in UAD models easier to estimate and aid in determining the most desirable dataset composition. FUNDING: European Research Council Deep4MI.

Machine learning analysis of human skin by optoacoustic mesoscopy for automated extraction of psoriasis and aging biomarkers.

Ultra-wideband raster-scan optoacoustic mesoscopy (RSOM) is a novel modality that has demonstrated unprecedented ability to visualize epidermal and dermal structures in-vivo. However, an automatic and quantitative analysis of three-dimensional RSOM datasets remains unexplored. In this work we present our framework: Deep Learning RSOM Analysis Pipeline (DeepRAP), to analyze and quantify morphological skin features recorded by RSOM and extract imaging biomarkers for disease characterization. DeepRAP uses a multi-network segmentation strategy based on convolutional neural networks with transfer learning. This strategy enabled the automatic recognition of skin layers and subsequent segmentation of dermal microvasculature with an accuracy equivalent to human assessment. DeepRAP was validated against manual segmentation on 25 psoriasis patients under treatment and our biomarker extraction was shown to characterize disease severity and progression well with a strong correlation to physician evaluation and histology. In a unique validation experiment, we applied DeepRAP in a time series sequence of occlusion-induced hyperemia from 10 healthy volunteers. We observe how the biomarkers decrease and recover during the occlusion and release process, demonstrating accurate performance and reproducibility of DeepRAP. Furthermore, we analyzed a cohort of 75 volunteers and defined a relationship between aging and microvascular features in-vivo. More precisely, this study revealed that fine microvascular features in the dermal layer have the strongest correlation to age. The ability of our newly developed framework to enable the rapid study of human skin morphology and microvasculature in-vivo promises to replace biopsy studies, increasing the translational potential of RSOM.

DWT-CNNTRN: a Convolutional Transformer for ECG Classification with Discrete Wavelet Transform.

Cardiovascular diseases are the leading cause of death worldwide. The diagnoses of cardiovascular diseases are usually carried out by cardiologists utilizing Electrocardiograms (ECGs). To assist these physicians in making an accurate diagnosis, there is a growing need for reliable and automatic ECG classifiers.In this study, a new method is proposed to classify 12-lead ECG recordings. The proposed model is composed of four components: the CNN(Convolutional Neural Network) module, the transformer module, the global hybrid pooling layer, and a classification layer. To improve the classification performance, the model takes the discrete wavelet transform of ECG signals as the model inputs and utilizes a hybrid pooling layer to condense the most important features over each period.The proposed model is evaluated using the test set of the China Physiological Signal Challenge 2018 dataset with 12-lead ECGs. It performs with an average accuracy of 0.86 and an average F1-scores of 0.83. The scores are particularly good for the block conditions (LBBB, RBBB, I-AVB). The main advantage of the proposed model is that, it obtains good results with a significantly smaller number of parameters compared to other individual and ensemble models.Clinical relevance- This work establishes a new ECG classifier model with high performance and low model size. It can make automatic ECG analysis more accessible, efficient, and accurate, especially in remote or underserved areas.

Combining generative modelling and semi-supervised domain adaptation for whole heart cardiovascular magnetic resonance angiography segmentation.

BACKGROUND: Quantification of three-dimensional (3D) cardiac anatomy is important for the evaluation of cardiovascular diseases. Changes in anatomy are indicative of remodeling processes as the heart tissue adapts to disease. Although robust segmentation methods exist for computed tomography angiography (CTA), few methods exist for whole-heart cardiovascular magnetic resonance angiograms (CMRA) which are more challenging due to variable contrast, lower signal to noise ratio and a limited amount of labeled data. METHODS: Two state-of-the-art unsupervised generative deep learning domain adaptation architectures, generative adversarial networks and variational auto-encoders, were applied to 3D whole heart segmentation of both conventional (n = 20) and high-resolution (n = 45) CMRA (target) images, given segmented CTA (source) images for training. An additional supervised loss function was implemented to improve performance given 10%, 20% and 30% segmented CMRA cases. A fully supervised nn-UNet trained on the given CMRA segmentations was used as the benchmark. RESULTS: The addition of a small number of segmented CMRA training cases substantially improved performance in both generative architectures in both standard and high-resolution datasets. Compared with the nn-UNet benchmark, the generative methods showed substantially better performance in the case of limited labelled cases. On the standard CMRA dataset, an average 12% (adversarial method) and 10% (variational method) improvement in Dice score was obtained. CONCLUSIONS: Unsupervised domain-adaptation methods for CMRA segmentation can be boosted by the addition of a small number of supervised target training cases. When only few labelled cases are available, semi-supervised generative modelling is superior to supervised methods.

Heartbeat Classification by Random Forest With a Novel Context Feature: A Segment Label.

OBJECTIVE: Physicians use electrocardiograms (ECG) to diagnose cardiac abnormalities. Sometimes they need to take a deeper look at abnormal heartbeats to diagnose the patients more precisely. The objective of this research is to design a more accurate heartbeat classification algorithm to assist physicians in identifying specific types of the heartbeat. METHODS AND PROCEDURES: In this paper, we propose a novel feature called a segment label, to improve the performance of a heartbeat classifier. This feature, provided by a Convolutional Neural Network, encodes the information surrounding the particular heartbeat. The random forest classifier is trained based on this new feature and other traditional features to classify the heartbeats. RESULTS: We validate our method on the MIT-BIH Arrhythmia dataset following the inter-patient evaluation paradigm. The proposed method is competitive with other similar works. It achieves an accuracy of 0.96, and F1-scores for normal beats, ventricular ectopic beats, and Supra-Ventricular Ectopic Beats (SVEB) of 0.98, 0.93, and 0.74, respectively. The precision and sensitivity for SVEB are 0.76 and 0.78, which outperforms the state-of-the-art methods. CONCLUSION: This study demonstrates that the segment label can contribute to precisely classifying heartbeats, especially those that require rhythm information as context information (e.g. SVEB). Clinical impact: Using a medical devices embedding our algorithm could ease the physicians' processes of diagnosing cardiovascular diseases, especially for SVEB, in clinical implementation.

Prediction of Complications and Surgery Duration in Primary Total Hip Arthroplasty Using Machine Learning: The Necessity of Modified Algorithms and Specific Data.

BACKGROUND: Machine Learning (ML) in arthroplasty is becoming more popular, as it is perfectly suited for prediction models. However, results have been heterogeneous so far. We hypothesize that an accurate ML model for outcome prediction in THA must be able to compute arthroplasty-specific data. In this study, we evaluate a ML approach applying data from two German arthroplasty-specific registries to predict adverse outcomes after THA, after careful evaluations of ML algorithms, outcome and input variables by an interdisciplinary team of data scientists and surgeons. METHODS: Data of 1217 cases of primary THA from a single center were derived from two German arthroplasty-specific registries between 2016 to 2019. The XGBoost algorithm was adjusted and applied. Accuracy, sensitivity, specificity and AUC were calculated. RESULTS: For the prediction of complications, the ML algorithm achieved an accuracy of 80.3%, a sensitivity of 31.0%, a specificity of 89.4% and an AUC of 64.1%. For the prediction of surgery duration, the ML algorithm yielded an accuracy of 81.7%, a sensitivity of 58.2%, a specificity of 91.6% and an AUC of 89.1%. The feature importance indicated non-linear outcomes for age, height, weight and surgeon. No relevant linear correlations were found. CONCLUSION: The attunement of input and output data as well as the modifications of the ML algorithm permitted the development of a feasible ML model for the prediction of complications and surgery duration.

Flimma: a federated and privacy-aware tool for differential gene expression analysis.

Aggregating transcriptomics data across hospitals can increase sensitivity and robustness of differential expression analyses, yielding deeper clinical insights. As data exchange is often restricted by privacy legislation, meta-analyses are frequently employed to pool local results. However, the accuracy might drop if class labels are inhomogeneously distributed among cohorts. Flimma ( https://exbio.wzw.tum.de/flimma/ ) addresses this issue by implementing the state-of-the-art workflow limma voom in a federated manner, i.e., patient data never leaves its source site. Flimma results are identical to those generated by limma voom on aggregated datasets even in imbalanced scenarios where meta-analysis approaches fail.

AI for Doctors-A Course to Educate Medical Professionals in Artificial Intelligence for Medical Imaging.

Successful adoption of artificial intelligence (AI) in medical imaging requires medical professionals to understand underlying principles and techniques. However, educational offerings tailored to the need of medical professionals are scarce. To fill this gap, we created the course "AI for Doctors: Medical Imaging". An analysis of participants' opinions on AI and self-perceived skills rated on a five-point Likert scale was conducted before and after the course. The participants' attitude towards AI in medical imaging was very optimistic before and after the course. However, deeper knowledge of AI and the process for validating and deploying it resulted in significantly less overoptimism with respect to perceivable patient benefits through AI (p = 0.020). Self-assessed skill ratings significantly improved after the course, and the appreciation of the course content was very positive. However, we observed a substantial drop-out rate, mostly attributed to the lack of time of medical professionals. There is a high demand for educational offerings regarding AI in medical imaging among medical professionals, and better education may lead to a more realistic appreciation of clinical adoption. However, time constraints imposed by a busy clinical schedule need to be taken into account for successful education of medical professionals.