The role of noise in denoising models for anomaly detection in medical images.

Pathological brain lesions exhibit diverse appearance in brain images, in terms of intensity, texture, shape, size, and location. Comprehensive sets of data and annotations are difficult to acquire. Therefore, unsupervised anomaly detection approaches have been proposed using only normal data for training, with the aim of detecting outlier anomalous voxels at test time. Denoising methods, for instance classical denoising autoencoders (DAEs) and more recently emerging diffusion models, are a promising approach, however naive application of pixelwise noise leads to poor anomaly detection performance. We show that optimization of the spatial resolution and magnitude of the noise improves the performance of different model training regimes, with similar noise parameter adjustments giving good performance for both DAEs and diffusion models. Visual inspection of the reconstructions suggests that the training noise influences the trade-off between the extent of the detail that is reconstructed and the extent of erasure of anomalies, both of which contribute to better anomaly detection performance. We validate our findings on two real-world datasets (tumor detection in brain MRI and hemorrhage/ischemia/tumor detection in brain CT), showing good detection on diverse anomaly appearances. Overall, we find that a DAE trained with coarse noise is a fast and simple method that gives state-of-the-art accuracy. Diffusion models applied to anomaly detection are as yet in their infancy and provide a promising avenue for further research. Code for our DAE model and coarse noise is provided at: https://github.com/AntanasKascenas/DenoisingAE.

Acute stroke CDS: automatic retrieval of thrombolysis contraindications from unstructured clinical letters.

Introduction: Thrombolysis treatment for acute ischaemic stroke can lead to better outcomes if administered early enough. However, contraindications exist which put the patient at greater risk of a bleed (e.g. recent major surgery, anticoagulant medication). Therefore, clinicians must check a patient's past medical history before proceeding with treatment. In this work we present a machine learning approach for accurate automatic detection of this information in unstructured text documents such as discharge letters or referral letters, to support the clinician in making a decision about whether to administer thrombolysis. Methods: We consulted local and national guidelines for thrombolysis eligibility, identifying 86 entities which are relevant to the thrombolysis decision. A total of 8,067 documents from 2,912 patients were manually annotated with these entities by medical students and clinicians. Using this data, we trained and validated several transformer-based named entity recognition (NER) models, focusing on transformer models which have been pre-trained on a biomedical corpus as these have shown most promise in the biomedical NER literature. Results: Our best model was a PubMedBERT-based approach, which obtained a lenient micro/macro F1 score of 0.829/0.723. Ensembling 5 variants of this model gave a significant boost to precision, obtaining micro/macro F1 of 0.846/0.734 which approaches the human annotator performance of 0.847/0.839. We further propose numeric definitions for the concepts of name regularity (similarity of all spans which refer to an entity) and context regularity (similarity of all context surrounding mentions of an entity), using these to analyse the types of errors made by the system and finding that the name regularity of an entity is a stronger predictor of model performance than raw training set frequency. Discussion: Overall, this work shows the potential of machine learning to provide clinical decision support (CDS) for the time-critical decision of thrombolysis administration in ischaemic stroke by quickly surfacing relevant information, leading to prompt treatment and hence to better patient outcomes.

Survey: Leakage and Privacy at Inference Time.

Leakage of data from publicly available Machine Learning (ML) models is an area of growing significance since commercial and government applications of ML can draw on multiple sources of data, potentially including users' and clients' sensitive data. We provide a comprehensive survey of contemporary advances on several fronts, covering involuntary data leakage which is natural to ML models, potential malicious leakage which is caused by privacy attacks, and currently available defence mechanisms. We focus on inference-time leakage, as the most likely scenario for publicly available models. We first discuss what leakage is in the context of different data, tasks, and model architectures. We then propose a taxonomy across involuntary and malicious leakage, followed by description of currently available defences, assessment metrics, and applications. We conclude with outstanding challenges and open questions, outlining some promising directions for future research.

Fully automated volumetric measurement of malignant pleural mesothelioma by deep learning AI: validation and comparison with modified RECIST response criteria.

BACKGROUND: In malignant pleural mesothelioma (MPM), complex tumour morphology results in inconsistent radiological response assessment. Promising volumetric methods require automation to be practical. We developed a fully automated Convolutional Neural Network (CNN) for this purpose, performed blinded validation and compared CNN and human response classification and survival prediction in patients treated with chemotherapy. METHODS: In a multicentre retrospective cohort study; 183 CT datasets were split into training and internal validation (123 datasets (80 fully annotated); 108 patients; 1 centre) and external validation (60 datasets (all fully annotated); 30 patients; 3 centres). Detailed manual annotations were used to train the CNN, which used two-dimensional U-Net architecture. CNN performance was evaluated using correlation, Bland-Altman and Dice agreement. Volumetric response/progression were defined as ≤30%/≥20% change and compared with modified Response Evaluation Criteria In Solid Tumours (mRECIST) by Cohen's kappa. Survival was assessed using Kaplan-Meier methodology. RESULTS: Human and artificial intelligence (AI) volumes were strongly correlated (validation set r=0.851, p<0.0001). Agreement was strong (validation set mean bias +31 cm3 (p=0.182), 95% limits 345 to +407 cm3). Infrequent AI segmentation errors (4/60 validation cases) were associated with fissural tumour, contralateral pleural thickening and adjacent atelectasis. Human and AI volumetric responses agreed in 20/30 (67%) validation cases κ=0.439 (0.178 to 0.700). AI and mRECIST agreed in 16/30 (55%) validation cases κ=0.284 (0.026 to 0.543). Higher baseline tumour volume was associated with shorter survival. CONCLUSION: We have developed and validated the first fully automated CNN for volumetric MPM segmentation. CNN performance may be further improved by enriching future training sets with morphologically challenging features. Volumetric response thresholds require further calibration in future studies.

Comparing performance on the Months of the Year Backwards test in hospitalised patients with delirium, dementia, and no cognitive impairment: an exploratory study.

PURPOSE: To investigate performance of the Months of the Year Backwards (MOTYB) test in older hospitalised patients with delirium, dementia, and no cognitive impairment. METHODS: Secondary analysis of data from a case-control study of 149 hospitalised patients aged ≥ 65 years with delirium [with or without dementia (N = 50)], dementia [without delirium (N = 46)], and no cognitive impairment (N = 53). Verbatim transcripts of MOTYB audio recordings were analysed to determine group differences in response patterns. RESULTS: In the total sample [median age 85y (IQR 80-88), 82% female], patients with delirium were more often unable to recite months backward to November (36/50 = 72%) than patients with dementia (21/46 = 46%; p < 0.01) and both differed significantly from patients without cognitive impairment (2/53 = 4%; p's < 0.001). 121/149 (81%) of patients were able to engage with the test. Patients with delirium were more often unable to engage with MOTYB (23/50 = 46%; e.g., due to reduced arousal) than patients with dementia (5/46 = 11%; p < 0.001); both groups differed significantly (p's < 0.001) from patients without cognitive impairment (0/53 = 0%). There was no statistically significant difference between patients with delirium (2/27 = 7%) and patients with dementia (8/41 = 20%) in completing MOTYB to January, but performance in both groups differed (p < 0.001 and p < 0.02, respectively) from patients without cognitive impairment (35/53 = 66%). CONCLUSION: Delirium was associated with inability to engage with MOTYB and low rates of completion. In patients able to engage with the test, error-free completion rates were low in delirium and dementia. Recording of engagement and patterns of errors may add useful information to MOTYB scoring.

A smartphone-based test for the assessment of attention deficits in delirium: A case-control diagnostic test accuracy study in older hospitalised patients.

BACKGROUND: Delirium is a common and serious acute neuropsychiatric syndrome which is often missed in routine clinical care. Inattention is the core cognitive feature. Diagnostic test accuracy (including cut-points) of a smartphone Delirium App (DelApp) for assessing attention deficits was assessed in older hospital inpatients. METHODS: This was a case-control study of hospitalised patients aged ≥65 years with delirium (with or without pre-existing cognitive impairment), who were compared to patients with dementia without delirium, and patients without cognitive impairment. Reference standard delirium assessment, which included a neuropsychological test battery, was based on Diagnostic and Statistical Manual of Mental Disorders-5 criteria. A separate blinded assessor administered the DelApp arousal assessment (score 0-4) and attention task (0-6) yielding an overall score of 0 to 10 (lower scores indicate poorer performance). Analyses included receiver operating characteristic curves and sensitivity and specificity. Optimal cut-points for delirium detection were determined using Youden's index. RESULTS: A total of 187 patients were recruited, mean age 83.8 (range 67-98) years, 152 (81%) women; n = 61 with delirium; n = 61 with dementia without delirium; and n = 65 without cognitive impairment. Patients with delirium performed poorly on the DelApp (median score = 4/10; inter-quartile range 3.0, 5.5) compared to patients with dementia (9.0; 5.5, 10.0) and those without cognitive impairment (10.0; 10.0, 10.0). Area under the curve for detecting delirium was 0.89 (95% Confidence Interval 0.84, 0.94). At an optimal cut-point of ≤8, sensitivity was 91.7% (84.7%, 98.7%) and specificity 74.2% (66.5%, 81.9%) for discriminating delirium from the other groups. Specificity was 68.3% (56.6%, 80.1%) for discriminating delirium from dementia (cut-point ≤6). CONCLUSION: Patients with delirium (with or without pre-existing cognitive impairment) perform poorly on the DelApp compared to patients with dementia and those without cognitive impairment. A cut-point of ≤8/10 is suggested as having optimal sensitivity and specificity. The DelApp is a promising tool for assessment of attention deficits associated with delirium in older hospitalised adults, many of whom have prior cognitive impairment, and should be further validated in representative patient cohorts.

Development and feasibility of a smartphone-based test for the objective detection and monitoring of attention impairments in delirium in the ICU.

PURPOSE: Delirium in the ICU is under-diagnosed. We evaluated feasibility and performance of a novel smartphone-based test for objectively detecting inattention in delirium. MATERIAL AND METHODS: DelApp-ICU combines a behavioural assessment and an attention task, whereby participants follow simple commands and count serially presented circles (score range 0-12, lower scores indicating worse performance). We assessed feasibility through staff interviews. Then we performed a preliminary case-control study in patients with and without delirium (ascertained with the Confusion Assessment Method for the ICU) who underwent the DelApp-ICU on up to 4 days. RESULTS: Forty-six patients (median age = 57.5 years, range 18-83) were assessed 89 times in total (N's = 46, 29, 10 and 4 for subsequent assessments; 33.7% delirious). DelApp-ICU scores were lower in delirium (N = 20; median = 0.5, Inter-Quartile Range (IQR) = 0-4.75) compared to no delirium (N = 26, median = 12, IQR = 8-12) on days 1, 2 and 3 (p < 0.001, p < 0.001 and p < 0.05, respectively). A DelApp-ICU score ≤6 was 100% sensitive and 96% specific to delirium on day 1. Positive and Negative Predictive Values were 91% and 100%, respectively. DelApp-ICU scores were responsive to changes in CAM-ICU status. CONCLUSIONS: DelApp-ICU shows promise for assessing inattention and delirium in ICU patients, including longitudinally monitoring deficits and providing a metric of delirium severity.

Diagnostic test accuracy of a novel smartphone application for the assessment of attention deficits in delirium in older hospitalised patients: a prospective cohort study protocol.

BACKGROUND: Delirium is a common and serious clinical syndrome which is often missed in routine clinical care. The core cognitive feature is inattention. We developed a novel bedside neuropsychological test for assessing inattention in delirium implemented on a smartphone platform (DelApp). We aim to evaluate the diagnostic performance of the DelApp in a representative cohort of older hospitalised patients. METHODS: This is a prospective study of older non-scheduled hospitalised patients (target n = 500, age ≥ 65), recruited from elderly care and acute orthopaedic wards. Exclusion criteria are: non-English speakers; severe vision or hearing impairment; photosensitive epilepsy. A structured reference standard delirium assessment based on DSM-5 criteria will be used, which includes a cognitive test battery administered by a trained assessor (Orientation-Memory-Concentration Test, Abbreviated Mental Test-10, Delirium Rating Severity Scale-Revised-98, digit span, months and days backwards, Vigilance A' test) and assessment of arousal (Observational Scale of Level of Arousal, Richmond Agitation Sedation Scale). Prior change in cognition will be documented using the Informant Questionnaire on Cognitive Decline in the Elderly. Patients will be categorized as delirium (with/without dementia), possible delirium, dementia, no cognitive impairment, or undetermined. A separate assessor (blinded to diagnosis and assessments) will administer the DelApp index test within 3 h of the reference standard assessment. The DelApp comprises assessment of arousal (score 0-4) and sustained attention (score 0-6), yielding a total score between 0 and 10 (higher score = better performance). Outcomes (length of stay, mortality and discharge location) will be collected at 12 weeks. We will evaluate a priori cutpoints derived from a previous case-control study. Measures of the accuracy of DelApp will include sensitivity, specificity, positive and negative predictive values, and area under the ROC curve. We plan repeat assessments on up to 4 occasions in a purposive subsample of 30 patients (15 delirium, 15 no delirium) to examine changes over time. DISCUSSION: This study evaluates the diagnostic test accuracy of a novel smartphone test for delirium in a representative cohort of older hospitalised patients, including those with dementia. DelApp has the potential to be a convenient, objective method of improving delirium assessment for older people in acute care. TRIAL REGISTRATION: Clinical trials.gov, NCT02590796 . Registered on 29 Oct 2015. Protocol version 5, dated 25 July 2016.

Accuracy and Feasibility of an Android-Based Digital Assessment Tool for Post Stroke Visual Disorders-The StrokeVision App.

BACKGROUND: Visual impairment affects up to 70% of stroke survivors. We designed an app (StrokeVision) to facilitate screening for common post stroke visual issues (acuity, visual fields, and visual inattention). We sought to describe the test time, feasibility, acceptability, and accuracy of our app-based digital visual assessments against (a) current methods used for bedside screening and (b) gold standard measures. METHODS: Patients were prospectively recruited from acute stroke settings. Index tests were app-based assessments of fields and inattention performed by a trained researcher. We compared against usual clinical screening practice of visual fields to confrontation, including inattention assessment (simultaneous stimuli). We also compared app to gold standard assessments of formal kinetic perimetry (Goldman or Octopus Visual Field Assessment); and pencil and paper-based tests of inattention (Albert's, Star Cancelation, and Line Bisection). Results of inattention and field tests were adjudicated by a specialist Neuro-ophthalmologist. All assessors were masked to each other's results. Participants and assessors graded acceptability using a bespoke scale that ranged from 0 (completely unacceptable) to 10 (perfect acceptability). RESULTS: Of 48 stroke survivors recruited, the complete battery of index and reference tests for fields was successfully completed in 45. Similar acceptability scores were observed for app-based [assessor median score 10 (IQR: 9-10); patient 9 (IQR: 8-10)] and traditional bedside testing [assessor 10 (IQR: 9-10); patient 10 (IQR: 9-10)]. Median test time was longer for app-based testing [combined time to completion of all digital tests 420 s (IQR: 390-588)] when compared with conventional bedside testing [70 s, (IQR: 40-70)], but shorter than gold standard testing [1,260 s, (IQR: 1005-1,620)]. Compared with gold standard assessments, usual screening practice demonstrated 79% sensitivity and 82% specificity for detection of a stroke-related field defect. This compares with 79% sensitivity and 88% specificity for StrokeVision digital assessment. CONCLUSION: StrokeVision shows promise as a screening tool for visual complications in the acute phase of stroke. The app is at least as good as usual screening and offers other functionality that may make it attractive for use in acute stroke. CLINICAL TRIAL REGISTRATION: https://ClinicalTrials.gov/ct2/show/NCT02539381.

Pre-clinical investigations of multi-path propagation in transcranial Doppler ultrasound flow phantom.

There are various practical situations in medical applications when pre-clinical investigations must be performed using a simulation environment or test bench prior to human studies. One example is the analysis of propagation channels in Transcanial Doppler (TCD) ultrasound (US), a signal processing challenge requiring the analysis of data from US waves scattered in three dimensions (3D). When examining the effects of scatterers in such channels, it is common to use a data acquisition test bench and a Doppler flow phantom. Such medical phantoms are frequently required to verify image and signal processing systems, and are often used to support algorithm development for a wide range of imaging and blood flow assessments. In this paper we describe a TCD simulation environment for the acquisition, investigation and pre-clinical data analysis of multi-path propagation in TCD US systems. This is demonstrated by comparing the anticipated theoretical and simulation channel statistics with the measured acoustic characteristics in terms of the probability distribution and autocorrelation functions.

Development of a smartphone application for the objective detection of attentional deficits in delirium.

BACKGROUND: Delirium is an acute, severe deterioration in mental functioning. Inattention is the core feature, yet there are few objective methods for assessing attentional deficits in delirium. We previously developed a novel, graded test for objectively detecting inattention in delirium, implemented on a computerized device (Edinburgh Delirium Test Box (EDTB)). Although the EDTB is effective, tests on universally available devices have potential for greater impact. Here we assessed feasibility and validity of the DelApp, a smartphone application based on the EDTB. METHODS: This was a preliminary case-control study in hospital inpatients (aged 60-96 years) with delirium (N = 50), dementia (N = 52), or no cognitive impairment (N = 54) who performed the DelApp assessment, which comprises an arousal assessment followed by counting of lights presented serially. Delirium was assessed using the Confusion Assessment Method and Delirium Rating Scale-Revised-98 (DRS-R98), and cognition with conventional tests of attention (e.g. digit span) and the short Orientation-Memory-Concentration Test (OMCT). RESULTS: DelApp scores (maximum score = 10) were lower in delirium (scores (median(IQR)): 6 (4-7)) compared to dementia (10 (9-10)) and control groups (10 (10-10), p-values < 0.001). Receiver operating characteristic (ROC) analyses revealed excellent accuracy of the DelApp for discriminating delirium from dementia (AUC = 0.93), and delirium from controls (AUC = 0.99, p-values < 0.001). DelApp and DRS-R98 severity scores were moderately well correlated (Kendall's tau = -0.60, p < 0.001). OMCT scores did not differ between delirium and dementia. CONCLUSIONS: The DelApp test showed good performance, supporting the utility of objectively measuring attention in delirium assessment. This study provides evidence of the feasibility of using a smartphone test for attentional assessment in hospital inpatients with possible delirium, with potential applications in research and clinical practice.

A wall-less poly(vinyl alcohol) cryogel flow phantom with accurate scattering properties for transcranial Doppler ultrasound propagation channels analysis.

Medical phantoms are frequently required to verify image and signal processing systems, and are often used to support algorithm development for a wide range of imaging and blood flow assessments. A phantom with accurate scattering properties is a crucial requirement when assessing the effects of multi-path propagation channels during the development of complex signal processing techniques for Transcranial Doppler (TCD) ultrasound. The simulation of physiological blood flow in a phantom with tissue and blood equivalence can be achieved using a variety of techniques. In this paper, poly (vinyl alcohol) cryogel (PVA-C) tissue mimicking material (TMM) is evaluated in conjunction with a number of potential scattering agents. The acoustic properties of the TMMs are assessed and an acoustic velocity of 1524ms(-1), an attenuation coefficient of (0:49) × 10(-4)fdBm(1)Hz(-1), a characteristic impedance of (1.72) × 10(6)Kgm(-2)s(-1) and a backscatter coefficient of (1.12) × 10(-28)f(4)m(-1)Hz(-4)sr(-1) were achieved using 4 freeze-thaw cycles and an aluminium oxide (Al(2)O(3)) scattering agent. This TMM was used to make an anatomically realistic wall-less flow phantom for studying the effects of multipath propagation in TCD ultrasound.

Waist belt and central obesity cause partial hiatus hernia and short-segment acid reflux in asymptomatic volunteers.

OBJECTIVE: There is a high incidence of inflammation and metaplasia at the gastro-oesophageal junction (GOJ) in asymptomatic volunteers. Additionally, the majority of patients with GOJ adenocarcinomas have no history of reflux symptoms. We report the effects of waist belt and increased waist circumference (WC) on the physiology of the GOJ in asymptomatic volunteers. DESIGN: 12 subjects with normal and 12 with increased WC, matched for age and gender were examined fasted and following a meal and with waist belts on and off. A magnet was clipped to the squamo-columnar junction (SCJ). Combined assembly of magnet-locator probe, 12-channel pH catheter and 36-channel manometer was passed. RESULTS: The waist belt and increased WC were each associated with proximal displacement of SCJ within the diaphragmatic hiatus (relative to upper border of lower oesophageal sphincter (LOS), peak LOS pressure point and pressure inversion point, and PIP (all p<0.05). The magnitude of proximal migration of SCJ during transient LOS relaxations was reduced by 1.6-2.6 cm with belt on versus off (p=0.01) and in obese versus non-obese (p=0.04), consistent with its resting position being already proximally displaced. The waist belt, but not increased WC, was associated with increased LOS pressure (vs intragastric pressure) and movement of pH transition point closer to SCJ. At 5 cm above upper border LOS, the mean % time pH <4 was <4% in all studied groups. Acid exposure 0.5-1.5 cm above SCJ was increased, with versus without, belt (p=0.02) and was most marked in obese subjects with belt. CONCLUSIONS: Our findings indicate that in asymptomatic volunteers, waist belt and central obesity cause partial hiatus herniation and short-segment acid reflux. This provides a plausible explanation for the high incidence of inflammation and metaplasia and occurrence of neoplasia at the GOJ in subjects without a history of reflux symptoms.

Development of Android apps for cognitive assessment of dementia and delirium.

The next generation of medical technology applications for hand-held portable platforms will provide a core change in performance and sophistication, transforming the way health care professionals interact with patients. This advance is particularly apparent in the delivery of cognitive patient assessments, where smartphones and tablet computers are being used to assess complex neurological conditions to provide objective, accurate and reproducible test results. This paper reports on two such applications (apps) that have been developed to assist healthcare professionals with the detection and diagnosis of dementia and delirium.

On the evolutionary and ontogenetic origins of tool-oriented behaviour in New Caledonian crows (Corvus moneduloides).

New Caledonian crows (Corvus moneduloides) are prolific tool users in captivity and in the wild, and have an inherited predisposition to express tool-oriented behaviours. To further understand the evolution and development of tool use, we compared the development of object manipulation in New Caledonian crows and common ravens (Corvus corax), which do not routinely use tools. We found striking qualitative similarities in the ontogeny of tool-oriented behaviour in New Caledonian crows and food-caching behaviour in ravens. Given that the common ancestor of New Caledonian crows and ravens was almost certainly a caching species, we therefore propose that the basic action patterns for tool use in New Caledonian crows may have their evolutionary origins in caching behaviour. Noncombinatorial object manipulations had similar frequencies in the two species. However, frequencies of object combinations that are precursors to functional behaviour increased in New Caledonian crows and decreased in ravens throughout the study period, ending 6 weeks post-fledging. These quantitative observations are consistent with the hypothesis that New Caledonian crows develop tool-oriented behaviour because of an increased motivation to perform object combinations that facilitate the necessary learning.

New Caledonian crows use tools for non-foraging activities.

Tool use is of great interest for cognitive research, largely because it can be particularly revealing about the underlying information processing mechanisms. Tool use that is inflexible or requires extensive experience to change, and that is only addressed towards specific targets such as food, is not likely to reflect unusual or particularly complex cognition. On the contrary, if tools are employed flexibly and for a variety of innovative purposes, then conventional combinations of inherited predispositions and associative learning are challenged and interesting questions emerge. Since New Caledonian crows (Corvus moneduloides) are especially adept at using and making tools for food extraction, we decided to examine their ability to generalise this to other contexts. We recorded how five pairs of New Caledonian crows interacted with novel objects that were not associated with food. We observed eight occasions in which the first contact with the novel object was mediated by a tool, suggesting that the function of the tool was for exploration. This is the first report of non-foraging tool use in New Caledonian crows, and it implies that the cognitive operations controlling tool-oriented behaviour in this species are more general than previously thought.

Infection of Harmonia axyridis (Coleoptera: Coccinellidae) by Hesperomyces virescens (Ascomycetes: Laboulbeniales): role of mating status and aggregation behavior.

The ectoparasitic fungus Hesperomyces virescens was studied on Harmonia axyridis in North Carolina, in the southeastern United States. A primary goal was to investigate transmission of the disease by examining the correlation between the pattern of fungal infection and seasonal change in host behavior. Beetles were collected as they arrived at their winter quarters at two sites; in one site they were also subsampled at mid- and late winter. Insects were sexed and weighed, fungal thalli were counted, and their location on the host body mapped; spermathecae of females were examined for sperm. Infection levels varied between sites, differed significantly between the sexes in one site but not the other, and increased by approximately 40% during winter. The distribution of thalli on the body changed seasonally, in concert with behavioral changes in the host. At fall flight, thalli were found most often on the posterior elytra of mated females, virgin females, and males. This is suggestive that the disease had been spread among both sexes via successful and failed copulation attempts; however, the relatively low incidence of infection on the male venter does not fit the sexual transmission scenario. During winter, thallus location shifts in concert with beetle aggregation behavior, with infections more often located on the head and legs. Fresh weight of beetles decreased by approx. 20% during winter, but was not affected by disease status. Prior to spring flight, uninfected females were preferred as mating partners, but the probable relationship between female age and infection status complicates interpretation of the data.