Feature blindness: A challenge for understanding and modelling visual object recognition.

Humans rely heavily on the shape of objects to recognise them. Recently, it has been argued that Convolutional Neural Networks (CNNs) can also show a shape-bias, provided their learning environment contains this bias. This has led to the proposal that CNNs provide good mechanistic models of shape-bias and, more generally, human visual processing. However, it is also possible that humans and CNNs show a shape-bias for very different reasons, namely, shape-bias in humans may be a consequence of architectural and cognitive constraints whereas CNNs show a shape-bias as a consequence of learning the statistics of the environment. We investigated this question by exploring shape-bias in humans and CNNs when they learn in a novel environment. We observed that, in this new environment, humans (i) focused on shape and overlooked many non-shape features, even when non-shape features were more diagnostic, (ii) learned based on only one out of multiple predictive features, and (iii) failed to learn when global features, such as shape, were absent. This behaviour contrasted with the predictions of a statistical inference model with no priors, showing the strong role that shape-bias plays in human feature selection. It also contrasted with CNNs that (i) preferred to categorise objects based on non-shape features, and (ii) increased reliance on these non-shape features as they became more predictive. This was the case even when the CNN was pre-trained to have a shape-bias and the convolutional backbone was frozen. These results suggest that shape-bias has a different source in humans and CNNs: while learning in CNNs is driven by the statistical properties of the environment, humans are highly constrained by their previous biases, which suggests that cognitive constraints play a key role in how humans learn to recognise novel objects.

Clarifying status of DNNs as models of human vision.

On several key issues we agree with the commentators. Perhaps most importantly, everyone seems to agree that psychology has an important role to play in building better models of human vision, and (most) everyone agrees (including us) that deep neural networks (DNNs) will play an important role in modelling human vision going forward. But there are also disagreements about what models are for, how DNN-human correspondences should be evaluated, the value of alternative modelling approaches, and impact of marketing hype in the literature. In our view, these latter issues are contributing to many unjustified claims regarding DNN-human correspondences in vision and other domains of cognition. We explore all these issues in this response.

68 Ga-DOTATATE PET/CT imaging in endogenous hyperinsulinemic hypoglycemia: A tertiary endocrine centre experience.

OBJECTIVE: Literature regarding utility of 68 Ga-DOTATATE PET/CT in insulinoma localization across various subgroups [benign/malignant/multiple endocrine neoplasia-1 (MEN-1) syndrome associated] remains scarce. In this study, the performance of 68 Ga-DOTATATE PET/CT was compared with contrast-enhanced computed tomography (CECT) and 68 Ga-NODAGA-Exendin-4 PET/CT (whenever available) in an endogenous hyperinsulinemic hypoglycemia (EHH) cohort. DESIGN: Retrospective audit. PATIENTS: EHH patients [N = 36, lesions (n) = 49, final diagnosis: benign sporadic insulinoma (BSI) (N = 20), malignant insulinoma (N = 4, n = 14), MEN-1 syndrome associated insulinoma (N = 9, n = 15), Munchausen syndrome (N = 2) and drug-induced hypoglycemia (N = 1)] having both preoperative imaging modalities (CECT and 68 Ga-DOTATATE PET/CT). MEASUREMENTS: Per-lesion sensitivity (Sn) and positive predictive value (PPV) for histopathological diagnosis of insulinoma. RESULTS: Sn and PPV of 68 Ga-DOTATATE PET/CT were 67.3% and 89.2%; 55% and 100%; 85.7% and 100%; and 66.7% and 77% for overall EHH, BSI, malignant, and MEN-1 syndrome associated insulinoma cohorts respectively. Despite having comparatively lower sensitivity in BSI cohort, 68 Ga-DOTATATE PET/CT localized a pancreatic tail lesion missed by other modalities. 68 Ga-DOTATATE PET/CT had comparatively higher sensitivity in malignant insulinoma than BSI cohort. 68 Ga-DOTATATE PET/CT also paved the way for successful response to 177 Lu-based peptide receptor radionuclide therapy (PRRT). In MEN-1 cases, lower PPV as compared with BSI was due to uptake in non-insulinoma pancreatic neuroendocrine tumours (Pan-NET). CONCLUSIONS: 68 Ga-DOTATATE PET/CT has supplemental role in selected cases of BSI with negative and/or discordant results with CECT and 68 Ga-NODAGA-Exendin-4 PET/CT. In malignant insulinoma, 68 Ga-DOTATATE-PET/CT has an additional theranostic potential. Interference due to uptake in non-insulinoma Pan-NET in MEN-1 syndrome may hinder insulinoma localization with 68 Ga-DOTATATE-PET/CT.

Clinical Spectrum of Adrenal Cushing's Syndrome and the Caution for Interpretation of Adrenocorticotrophic Hormone: A Single-Center Experience.

To describe the differences in presentation, biochemistry, and radiological evaluation of various etiologies of adrenal Cushing's syndrome (CS) from a single center. To emphasize caution for interpretation of plasma adrenocorticotropic hormone (ACTH), as a spuriously unsuppressed ACTH level by immunometric assay may lead to therapeutic misadventures in adrenal CS. DESIGN: Retrospective, single-center, observational study. METHODS: Fifty-eight adrenal CS patients [Adrenocortical carcinoma (ACC), n=30; Adenoma (ACA), n=15; Primary pigmented nodular adrenocortical disease (PPNAD), n=10; ACTH independent macronodular adrenal hyperplasia (AIMAH), n=3) evaluated at a tertiary care center in western India between January 2006 to March 2020 were included. Data on demography, clinical evaluation, biochemistry, imaging, management, histopathology, and outcome were recorded in a standard format and analyzed. RESULTS: Cortisol secreting ACC presented at 38(1-50) years with abdominal mass in 26/30 (86.7%) and 16/30 (53.3%) had metastases at presentation. ACA with autonomous cortisol excess presented at 25(4.9-40) years with discriminating features of CS in 14/15 (93.3%), sex steroid production in 2/15, unenhanced HU <10 in only one, and relative washout >40% in 8/11 (72.7%). One ACA and eight ACC patients had plasma ACTH (by Siemens Immulite assay) > 20 pg/ml, despite hypercortisolemic state. CONCLUSIONS: Cortisol-secreting ACC and ACA most often present with mass effects and florid CS, respectively. Baseline HU has low sensitivity to differentiate cortisol-secreting ACA from ACC. Plasma ACTH measured by Seimens Immulite is often unsuppressed, especially in ACC patients, which can be addressed by measuring ACTH by more accurate assays.

Deep problems with neural network models of human vision.

Deep neural networks (DNNs) have had extraordinary successes in classifying photographic images of objects and are often described as the best models of biological vision. This conclusion is largely based on three sets of findings: (1) DNNs are more accurate than any other model in classifying images taken from various datasets, (2) DNNs do the best job in predicting the pattern of human errors in classifying objects taken from various behavioral datasets, and (3) DNNs do the best job in predicting brain signals in response to images taken from various brain datasets (e.g., single cell responses or fMRI data). However, these behavioral and brain datasets do not test hypotheses regarding what features are contributing to good predictions and we show that the predictions may be mediated by DNNs that share little overlap with biological vision. More problematically, we show that DNNs account for almost no results from psychological research. This contradicts the common claim that DNNs are good, let alone the best, models of human object recognition. We argue that theorists interested in developing biologically plausible models of human vision need to direct their attention to explaining psychological findings. More generally, theorists need to build models that explain the results of experiments that manipulate independent variables designed to test hypotheses rather than compete on making the best predictions. We conclude by briefly summarizing various promising modeling approaches that focus on psychological data.

Exendin-4-based imaging in insulinoma localization: Systematic review and meta-analysis.

BACKGROUND AND CONTEXT: Glucagon-like peptide-1 receptor (GLP-1 R) based imaging has shown higher sensitivity for insulinoma localization as compared to other anatomic/functional imaging. METHODOLOGY: We reviewed the published English literature for GLP-1 R targeted imaging in insulinoma in PubMed until August 2020 in accordance with PRISMA guidelines using the MeSH terms "((Exendin-4 PET/CT) OR (Exendin-4 SPECT/CT) OR (GLP-1 R imaging)) AND (Insulinoma)". An individual patient data-metanalysis (IPD-MA) was performed, and performance parameters were calculated for the histopathological diagnosis of insulinoma. MAIN OUTCOME MEASURES: True-positive (TP), false-positive (FP), false-negative (FN), true-negative (TN), sensitivity (Sn), specificity (Sp), positive predictive value (PPV) and negative predictive value (NPV) for insulinoma localization. RESULTS: A total of 179 cases (316 lesions) from 16 publications were included for IPD-MA. For insulinoma localization, exendin-4-PET/CT (Sn & PPV: 94%) performed better than exendin-4-SPECT/CT (Sn: 63%, PPV: 94%). The Sn was lower in malignant insulinoma cases whereas the Sp was higher in cases with MEN-1 syndrome. With exendin-4-based imaging, FP uptakes in Brunner's gland, normal pancreas, and other ß-cell pathologies and FN results in pancreatic tail lesions and malignancy were seen in a few patients. TN results suggested the correct diagnosis of other endogenous hyperinsulinemic hypoglycaemia (EHH) subtypes. CONCLUSION: For insulinoma localization, exendin-4 PET/CT should be preferred over exendin-4 SPECT/CT because of higher sensitivity and specificity. FP uptakes in Brunner's gland, normal pancreas, and other ß-cell pathologies and FN results in tail lesions, and malignant insulinomas are limitations. Higher specificity for insulinoma localization is particularly useful in patients with MEN-1 syndrome.

Assisted Fluid Management Software Guidance for Intraoperative Fluid Administration.

BACKGROUND: Excessive or inadequate fluid administration causes complications, but despite this, fluid administration during noncardiac surgery is highly variable. Goal-directed management helps optimize the amount and timing of fluid administration; however, implementation is difficult because algorithms are complex. The authors therefore tested the performance of the Acumen Assisted Fluid Management software (Edwards Lifesciences, USA), which is designed to guide optimal intravenous fluid administration during surgery. METHODS: In this multicenter, prospective, single-arm cohort evaluation, the authors enrolled 330 adults scheduled for moderate- to high-risk noncardiac surgery that required arterial catheter insertion and mechanical ventilation. Clinicians chose a fluid strategy based on a desired 10%, 15%, or 20% increase in stroke volume (SV) in response to a fluid bolus. Dedicated fluid management software prompted "test" or "recommended" boluses, and clinicians were free to initiate a "user" bolus of 100 to 500 ml of crystalloid or colloid. Clinicians were free to accept or decline the software prompts. The authors primarily compared the fraction of software-recommended boluses that produced suitable increases in SV to a 30% reference rate. On an exploratory basis, we compared responses to software-recommended and clinician-initiated boluses. RESULTS: Four hundred twenty-four of 479 (89%) software-recommended fluid boluses and 508 of 592 (86%) clinician-initiated fluid boluses were analyzed per protocol. Of those, 66% (95% CI, 62 to 70%) of delivered fluid boluses recommended by the software resulted in desired increases in SV, compared with the 30% reference rate, whereas only 41% (95% CI, 38 to 44%) of clinician-initiated boluses did (P < 0.0001). The mean ± SD increase in SV after boluses recommended by the software was 14.2 ± 13.9% versus 8.3 ± 12.1% (P < 0.0001) for those initiated by clinicians. CONCLUSIONS: Fluid boluses recommended by the software resulted in desired SV increases more often, and with greater absolute SV increase, than clinician-initiated boluses. Automated assessment of fluid responsiveness may help clinicians optimize intraoperative fluid management during noncardiac surgery.

Exendin-4-based imaging in endogenous hyperinsulinemic hypoglycaemia cohort: A tertiary Endocrine centre experience.

CONTEXT: Insulinoma needs accurate preoperative localization for minimally invasive surgery. Exendin-4-based imaging has shown promising results. OBJECTIVE: To evaluate performance parameters of exendin-4-based imaging in insulinoma localization and compare with other imaging modalities. DESIGN: Retrospective cross-sectional study. PATIENTS: We report 14 patients with endogenous hyperinsulinemic hypoglycaemia (EHH) managed at our centre; in whom, the final diagnosis was insulinoma (n = 11), Munchausen syndrome (MS) (n = 2) and inconclusive (n = 1). Retrospective reporting of CECT, 68 Ga-DOTATATE PET/CT and 68 Ga-NODAGA-exendin-4-PET/CT was done. With per-lesion analysis, performance parameters were calculated for the histopathological diagnosis of insulinoma. MAIN OUTCOME MEASURES: True positive (TP), false positive (FP), false negative (FN), true negative (TN), sensitivity (Sn), specificity (Sp), positive predictive value (PPV), negative predictive value (NPV) for insulinoma localization. RESULTS: In our cohort, 12 histopathologically proven insulinoma lesions [(TP): 11 primary lesions, 1 metastasis] were detected in 11 patients, whereas two patients had MS (TN). Sn and PPV were 75% and 100%, 33.3% and 80% and 83.3% and 71.4% for CECT, 68 Ga-DOTATATE PET/CT and 68 Ga-NODAGA-exendin-4-PET/CT, respectively. With exendin-4-based imaging, FP uptake in normal pancreatic tissue and FN results in the pancreatic tail lesion was seen. In one patient, TN result suggested the correct diagnosis of MS. CONCLUSION: 68 Ga-NODAGA-exendin-4-PET/CT has higher sensitivity than 68 Ga-DOTATATE PET/CT and CECT for insulinoma localization. FP uptake in normal pancreas and FN result in tail lesions are limitations of currently utilized exendin-4-based imaging.

Mechanistic models must link the field and the lab.

In the theory outlined in the target article, an animal forages continuously, making sequential decisions in a world where the amount of food and its uncertainty are fixed, but delays are variable. These assumptions contrast with the risk-sensitive foraging theory and create a problem for comparing the predictions of this model with many laboratory experiments that do not make these assumptions.

Optimal gut size of small birds and its dependence on environmental and physiological parameters.

Most optimal foraging models assume that the foraging behaviour of small birds depends on a single state variable, their energy reserves in the form of stored fat. Here, we include a second state variable-the contents of the bird's gut-to investigate how a bird should optimise its gut size to minimise its long-term mortality, depending on the availability of food, the size of meal and the bird's digestive constraints. Our results show that (1) the current level of fat is never less important than gut contents in determining the bird's survival; (2) there exists a unique optimal gut size, which is determined by a trade-off between the energetic gains and costs of maintaining a large digestive system; (3) the optimal gut size increases as the bird's digestive cycle becomes slower, allowing the bird to store undigested food; (4) the critical environmental factor for determining the optimal gut size is the mass of food found in a successful foraging effort ("meal size"). We find that when the environment is harsh, it is optimal for the bird to maintain a gut that is larger than the size of a meal. However, the optimal size of the gut in rich environments exactly matches the meal size (i.e. the mass of food that the optimal gut can carry is exactly the mass of food that can be obtained in a successful foraging attempt).

PRIMARY ADRENAL LYMPHOMA: A SINGLE-CENTER EXPERIENCE.

OBJECTIVE: To describe the clinical presentation, biochemistry, imaging features, and treatment outcome of patients with primary adrenal lymphoma (PAL) presenting to a single tertiary care center. METHODS: We performed a retrospective analysis of case records of 7 patients diagnosed with PAL between January 2011 and May 2014 at our institution in Mumbai, India. RESULTS: Median age of presentation in our series was 48 years (range, 41 to 60 years), with a male to female ratio of 6:1. Bilateral adrenal involvement was seen in 4 of 7 patients (58%). Adrenal insufficiency (AI) was seen in 3 of the 4 patients with bilateral involvement (75%). Computed tomography showed slight to moderate contrast enhancement of adrenal masses in 4 of 5 patients (80%). Diffuse, large, B-cell lymphoma (DLBCL) was the most common immunophenotype (85%). One patient died due to rapid disease progression even before starting chemotherapy. Six patients were treated with chemotherapy and/or external beam radiotherapy. After 1 year, 2 more patients had died, whereas 4 patients were in remission. CONCLUSION: PAL should always be considered in differential diagnosis of bilateral adrenal mass with AI. DLBCL is the most common histologic subtype of PAL. Despite treatment, long-term prognosis of PAL remains poor.

Functional imaging in primary tumour-induced osteomalacia: relative performance of FDG PET/CT vs somatostatin receptor-based functional scans: a series of nine patients.

CONTEXT: Localization of phosphatonin-producing mesenchymal tumours in patients with primary tumour-induced osteomalacia (pTIO) is challenging. Functional imaging plays an important role in the localization of these tumours. OBJECTIVE: We studied the relative performance of different functional imaging modalities ((18) F-FDG PET/CT, (99) Tc-HYNIC-TOC SPECT/CT and (68) Ga-DOTATATE PET/CT) in tumour localization in cases of pTIO. DESIGN AND METHODS: Retrospective chart evaluation of 16 patients with confirmed TIO treated from 2006 to 2013 was conducted in a tertiary care referral centre. RESULTS: Of 16, nine patients had pTIO. In these nine, the positivity rates of different functional imaging modalities were 50% for (18) F-FDG PET/CT (four of eight patients), 100% for (99) Tc-HYNIC-TOC SPECT/CT (six of six patients) and 100% for (68) Ga-DOTATATE PET/CT (seven of seven patients). Of nine patients, six were subjected to both the (99) Tc-HYNIC-TOC SPECT/CT and (68) Ga-DOTATATE PET/CT and all of them showed coregistration on the two scans. CONCLUSIONS: In patients with pTIO, the somatostatin receptor-based functional scans performed better than (18) F-FDG PET/CT in tumour localization. Amongst the somatostatin receptor-based scans, (99) Tc-HYNIC-TOC SPECT/CT and (68) Ga-DOTATATE PET/CT performed equally well for localization of tumours.

Rare Presentations of Differentiated Thyroid Cancer Exposing Dr Jekyll and Mr Hyde Nature of an Otherwise Indolent Disease: Case Series.

ABSTRACT: Differentiated thyroid carcinoma (DTC) usually manifests as an indolent cancer with good prognosis. However, rarely uncommon sites of metastatic involvement can worsen the prognosis and require aggressive therapeutic approach. Here in, we describe 5 patients (3 women and 2 men) harboring rare sites of metastatic involvement from DTC including the adrenals, colon, kidneys, urinary bladder, brachial plexus, and superior vena cava with contiguous right atrial involvement. The awareness of such rare sites of involvement from DTC is imperative for treating clinicians to plan individualistic approach in management including multiprong therapies for better patient care.

AcumenTM hypotension prediction index guidance for prevention and treatment of hypotension in noncardiac surgery: a prospective, single-arm, multicenter trial.

BACKGROUND: Intraoperative hypotension is common during noncardiac surgery and is associated with postoperative myocardial infarction, acute kidney injury, stroke, and severe infection. The Hypotension Prediction Index software is an algorithm based on arterial waveform analysis that alerts clinicians of the patient's likelihood of experiencing a future hypotensive event, defined as mean arterial pressure < 65 mmHg for at least 1 min. METHODS: Two analyses included (1) a prospective, single-arm trial, with continuous blood pressure measurements from study monitors, compared to a historical comparison cohort. (2) A post hoc analysis of a subset of trial participants versus a propensity score-weighted contemporaneous comparison group, using external data from the Multicenter Perioperative Outcomes Group (MPOG). The trial included 485 subjects in 11 sites; 406 were in the final effectiveness analysis. The post hoc analysis included 457 trial participants and 15,796 comparison patients. Patients were eligible if aged 18 years or older, American Society of Anesthesiologists (ASA) physical status 3 or 4, and scheduled for moderate- to high-risk noncardiac surgery expected to last at least 3 h. MEASUREMENTS: minutes of mean arterial pressure (MAP) below 65 mmHg and area under MAP < 65 mmHg. RESULTS: Analysis 1: Trial subjects (n = 406) experienced a mean of 9 ± 13 min of MAP below 65 mmHg, compared with the MPOG historical control mean of 25 ± 41 min, a 65% reduction (p < 0.001). Subjects with at least one episode of hypotension (n = 293) had a mean of 12 ± 14 min of MAP below 65 mmHg compared with the MPOG historical control mean of 28 ± 43 min, a 58% reduction (p< 0.001). Analysis 2: In the post hoc inverse probability treatment weighting model, patients in the trial demonstrated a 35% reduction in minutes of hypotension compared to a contemporaneous comparison group [exponentiated coefficient: - 0.35 (95%CI - 0.43, - 0.27); p < 0.001]. CONCLUSIONS: The use of prediction software for blood pressure management was associated with a clinically meaningful reduction in the duration of intraoperative hypotension. Further studies must investigate whether predictive algorithms to prevent hypotension can reduce adverse outcomes. TRIAL REGISTRATION: Clinical trial number: NCT03805217. Registry URL: https://clinicaltrials.gov/ct2/show/NCT03805217 . Principal investigator: Xiaodong Bao, MD, PhD. Date of registration: January 15, 2019.

Human shape representations are not an emergent property of learning to classify objects.

Humans are particularly sensitive to relationships between parts of objects. It remains unclear why this is. One hypothesis is that relational features are highly diagnostic of object categories and emerge as a result of learning to classify objects. We tested this by analyzing the internal representations of supervised convolutional neural networks (CNNs) trained to classify large sets of objects. We found that CNNs do not show the same sensitivity to relational changes as previously observed for human participants. Furthermore, when we precisely controlled the deformations to objects, human behavior was best predicted by the number of relational changes while CNNs were equally sensitive to all changes. Even changing the statistics of the learning environment by making relations uniquely diagnostic did not make networks more sensitive to relations in general. Our results show that learning to classify objects is not sufficient for the emergence of human shape representations. Instead, these results suggest that humans are selectively sensitive to relational changes because they build representations of distal objects from their retinal images and interpret relational changes as changes to these distal objects. This inferential process makes human shape representations qualitatively different from those of artificial neural networks optimized to perform image classification. (PsycInfo Database Record (c) 2023 APA, all rights reserved).

The role of capacity constraints in Convolutional Neural Networks for learning random versus natural data.

Convolutional neural networks (CNNs) are often described as promising models of human vision, yet they show many differences from human abilities. We focus on a superhuman capacity of top-performing CNNs, namely, their ability to learn very large datasets of random patterns. We verify that human learning on such tasks is extremely limited, even with few stimuli. We argue that the performance difference is due to CNNs' overcapacity and introduce biologically inspired mechanisms to constrain it, while retaining the good test set generalisation to structured images as characteristic of CNNs. We investigate the efficacy of adding noise to hidden units' activations, restricting early convolutional layers with a bottleneck, and using a bounded activation function. Internal noise was the most potent intervention and the only one which, by itself, could reduce random data performance in the tested models to chance levels. We also investigated whether networks with biologically inspired capacity constraints show improved generalisation to out-of-distribution stimuli, however little benefit was observed. Our results suggest that constraining networks with biologically motivated mechanisms paves the way for closer correspondence between network and human performance, but the few manipulations we have tested are only a small step towards that goal.

Angiogenesis Imaging of Adrenocortical Carcinoma with Ga-68-NODAGA-RGD Positron Emission Tomography: Opening New Horizons in Multimodality Imaging from Theranostic Perspective.

A 53-year-old female, with a known case of adrenocortical carcinoma (ACC), underwent F-18 fluorodeoxyglucose (FDG) positron emission tomography/computed tomography (PET/CT) for initial staging, which revealed FDG avid large left suprarenal mass contiguous with hypermetabolic tumor thrombus in the inferior vena cava (IVC) through the left renal vein. Thereafter, she underwent angiogenesis imaging using Ga-68-NODAGA-RGD PET/CT, which showed similar avid tracer uptake in both primary and IVC thrombus. Demonstration of RGD avidity in ACC in this case opens a new horizon for targeted radionuclide therapy (e.g., Lu-177 RGD) in selected patients, who may have limited therapeutic options.

Paediatric and adolescent ectopic Cushing's syndrome: systematic review.

OBJECTIVE: The data on clinical, biochemical, radiological characteristics, and outcomes in paediatric ectopic adrenocorticotropic hormone syndrome (EAS) are limited owing to rarity of the condition. We report three new cases and perform a systematic review of paediatric EAS. DESIGN AND METHOD: Case records of paediatric and adolescent EAS patient's ≤20 years presenting at our centre between 1997 and 2021 were retrospectively reviewed, and a systematic review of the literature published between January 1970 and December 2022 was performed. RESULTS: A total of 161 patients including 3 new patients from our centre were identified. Bronchial neuroendocrine tumours (NET) (28.5%), thymic NET (22.9%), primitive cell-derived tumours (18.6%), and gastro-entero-pancreatic-NET (13.7%) were the common causes. Primitive cell-derived tumours were the most common in the first decade (24/45, 53.4%) and were the largest (82 [60-100] mm), whereas bronchial NETs predominated during the second decade (42/116, 36.2%) and were the smallest (15 [10-25] mm). Computed tomography localized 92.9% (118/127) of paediatric EAS patients. Immediate postoperative remission was attained in 77.9% (88/113) patients, whereas 30.4% (24/79) relapsed over a median (IQR) period of 13 (8-36) months. Over a median (IQR) follow-up of 2 (0.6-4.6) years, 31.4% of patients died. The median survival was higher in bronchial NET than in other tumour groups. Distant metastasis and tumour size were independent negative predictors of survival. CONCLUSIONS: Aetiological profile of paediatric and adolescent EAS is distinct from that of adults. Bronchial NETs have the best long-term survival, whereas distant metastasis and tumour size predict poor survival.

Biological convolutions improve DNN robustness to noise and generalisation.

Deep Convolutional Neural Networks (DNNs) have achieved superhuman accuracy on standard image classification benchmarks. Their success has reignited significant interest in their use as models of the primate visual system, bolstered by claims of their architectural and representational similarities. However, closer scrutiny of these models suggests that they rely on various forms of shortcut learning to achieve their impressive performance, such as using texture rather than shape information. Such superficial solutions to image recognition have been shown to make DNNs brittle in the face of more challenging tests such as noise-perturbed or out-of-distribution images, casting doubt on their similarity to their biological counterparts. In the present work, we demonstrate that adding fixed biological filter banks, in particular banks of Gabor filters, helps to constrain the networks to avoid reliance on shortcuts, making them develop more structured internal representations and more tolerance to noise. Importantly, they also gained around 20-35% improved accuracy when generalising to our novel out-of-distribution test image sets over standard end-to-end trained architectures. We take these findings to suggest that these properties of the primate visual system should be incorporated into DNNs to make them more able to cope with real-world vision and better capture some of the more impressive aspects of human visual perception such as generalisation.

Suicidal behaviour: What's the brain up to?

BACKGROUND: Individuals with suicidal behaviours are increasingly recognized as having impairments in brain metabolism. However, these are not well delineated. AIM: To evaluate regional cerebral glucose metabolism (rCMglu) in subjects with suicidal behaviours and assess differences in rCMglu between depressed and non-depressed suicidal subjects. METHODS: Thirty-three subjects with suicidal behaviours were assessed using Columbia Suicide Severity Rating scale (CSSRS) and Beck's Depression Inventory (BDI). Brain metabolism was assessed using [18F]Fluoro,Deoxy-Glucose Positron Emission Tomography (FDG-PET). RESULTS: Of 33 subjects, eighteen had depression. FDG-PET findings revealed that in comparison to mean asymptomatic controls, subjects had decreased rCMglu in right inferior frontal, left Broca's, left inferiolateral andsuperiolateral temporal, right inferior parietal and left posterior cingulate cortex. Increased rCMglu was seen in bilateral superior and medial frontal, right inferiolateral and posteriomedial temporal cortex, and midbrain. CSSRS total intensity inversely correlated with rCMglu in medial frontal cortex, left Broca's and superiolateral temporal cortex and directly correlated with rCMglu in right cerebellum. There was no significant difference in rCMglu between depressed and non depressed group. CONCLUSIONS: Significant differences exist in rCMglu of suicidal individuals, chiefly in frontal and temporal regions. Understanding these would help us identify individuals more at risk for suicidal behaviours.