Lowering the computational barrier: Partially Bayesian neural networks for transparency in medical imaging AI

Deep Neural Networks (DNNs) can provide clinicians with fast and accurate predictions that are highly valuable for high-stakes medical decision-making, such as in brain tumor segmentation and treatment planning. However, these models largely lack transparency about the uncertainty in their predictions, potentially giving clinicians a false sense of reliability that may lead to grave consequences in patient care. Growing calls for Transparent and Responsible AI have promoted Uncertainty Quantification (UQ) to capture and communicate uncertainty in a systematic and principled manner. However, traditional Bayesian UQ methods remain prohibitively costly for large, million-dimensional tumor segmentation DNNs such as the U-Net. In this work, we discuss a computationally-efficient UQ approach via the partially Bayesian neural networks (pBNN). In pBNN, only a single layer, strategically selected based on gradient-based sensitivity analysis, is targeted for Bayesian inference. We illustrate the effectiveness of pBNN in capturing the full uncertainty for a 7.8-million parameter U-Net. We also demonstrate how practitioners and model developers can use the pBNN's predictions to better understand the model's capabilities and behavior.

Federated learning enables big data for rare cancer boundary detection

Although machine learning (ML) has shown promise across disciplines, out-of-sample generalizability is concerning. This is currently addressed by sharing multi-site data, but such centralization is challenging/infeasible to scale due to various limitations. Federated ML (FL) provides an alternative paradigm for accurate and generalizable ML, by only sharing numerical model updates. Here we present the largest FL study to-date, involving data from 71 sites across 6 continents, to generate an automatic tumor boundary detector for the rare disease of glioblastoma, reporting the largest such dataset in the literature (n = 6, 314). We demonstrate a 33% delineation improvement for the surgically targetable tumor, and 23% for the complete tumor extent, over a publicly trained model. We anticipate our study to: 1) enable more healthcare studies informed by large diverse data, ensuring meaningful results for rare diseases and underrepresented populations, 2) facilitate further analyses for glioblastoma by releasing our consensus model, and 3) demonstrate the FL effectiveness at such scale and task-complexity as a paradigm shift for multi-site collaborations, alleviating the need for data-sharing.

P-114 INVESTIGATING THE ROLE OF BOTOX IN THE EMERGENCY MANAGEMENT OF COMPLEX ABDOMINAL WALL HERNIAS

Aims

Botulinum toxin (Botox) has an established role in the elective management of complex abdominal wall hernia, but its role in the emergency setting has not been investigated. Clinical effects of Botox peak at 2–3 weeks after injection, but effects become apparent within 48 hours. We aimed to determine if these early effects can facilitate the acute reduction of irreducible hernia.

Materials and methods

Patients with irreducible abdominal wall hernias over a 30 month period were considered for Botox therapy. Suitable patients were given 300 units of Botox under ultrasound guidance into the anterolateral muscles bilaterally. Data were collected on outcomes.

Results

Twelve patients received Botox. All had incisional hernias and an ASA of 3 or more. Six were on antiplatelets and/or anticoagulation. All received Botox with standard conservative management. In ten patients, the hernia was reduced and symptoms resolved. One patient required surgery at 48 hours due to bowel ischaemia from inter-sac adhesions. A second patient experienced recurrent obstruction despite successful hernia reduction but was palliated due to advanced age and frailty.

Conclusions

This case series is the first study to investigate the role of Botox in the acute management of irreducible hernias. Our findings suggest that in selected patients, Botox can aid hernia reduction and avoid the morbidity of emergency repair. This is of particular value in high risk patients and those with complex defects who benefit from preoperative optimisation techniques. The accurate identification of strangulation is crucial however, as it excludes patients from this approach.

Stratification by Tumor Grade Groups in a Holistic Evaluation of Machine Learning for Brain Tumor Segmentation

Accurate and consistent segmentation plays an important role in the diagnosis, treatment planning, and monitoring of both High Grade Glioma (HGG), including Glioblastoma Multiforme (GBM), and Low Grade Glioma (LGG). Accuracy of segmentation can be affected by the imaging presentation of glioma, which greatly varies between the two tumor grade groups. In recent years, researchers have used Machine Learning (ML) to segment tumor rapidly and consistently, as compared to manual segmentation. However, existing ML validation relies heavily on computing summary statistics and rarely tests the generalizability of an algorithm on clinically heterogeneous data. In this work, our goal is to investigate how to holistically evaluate the performance of ML algorithms on a brain tumor segmentation task. We address the need for rigorous evaluation of ML algorithms and present four axes of model evaluation-diagnostic performance, model confidence, robustness, and data quality. We perform a comprehensive evaluation of a glioma segmentation ML algorithm by stratifying data by specific tumor grade groups (GBM and LGG) and evaluate these algorithms on each of the four axes. The main takeaways of our work are-(1) ML algorithms need to be evaluated on out-of-distribution data to assess generalizability, reflective of tumor heterogeneity. (2) Segmentation metrics alone are limited to evaluate the errors made by ML algorithms and their describe their consequences. (3) Adoption of tools in other domains such as robustness (adversarial attacks) and model uncertainty (prediction intervals) lead to a more comprehensive performance evaluation. Such a holistic evaluation framework could shed light on an algorithm's clinical utility and help it evolve into a more clinically valuable tool.

Everolimus plus octreotide long-acting repeatable in advanced neuroendocrine tumors in the routine tertiary cancer care setting: An Indian experience

BACKGROUND

Neuroendocrine tumors (NETs) are rare, heterogeneous, indolent tumors that are relatively insensitive to systemic chemotherapy. Therapeutic strategies for NETs broadly include somatostatin analogs, antiangiogenic therapy, and most recently, mammalian target of rapamycin inhibition. Combination therapy has shown promising antitumor activity and good tolerability in the randomized phase III trials.

AIM

The aim was to evaluate the safety and efficacy of Everolimus plus Octreotide long-acting repeatable (LAR) in patients with advanced NETs in the routine tertiary cancer care setting in India in this postapproval, noninterventional trial.

PATIENTS AND METHODS

Patients presenting to selected centers between 2011 and 2013 with histologically confirmed low-, intermediate- or high-grade advanced NETs who may have had prior exposure to cytotoxic chemotherapy (≤2 lines) were treated with oral Everolimus (10 mg/day) plus intramuscular Octreotide LAR (30 mg once every 28 days) until disease progression or unacceptable toxicity was seen. Patients were evaluated every 3 months for a response to therapy as per Response Evaluation Criteria in Solid Tumors.

RESULTS

Everolimus plus Octreotide LAR was associated with a clinical benefit rate of 69% (best evaluable responses: Stable disease [SD] in 10 patients [63%], partial response in 1 patient [6%]). The average duration of therapy was 4.8 cycles, and 3 (17%) patients continued therapy for ≥12 cycles (all achieved SD). The therapy was found to be well-tolerated in all patients.

CONCLUSIONS

Everolimus plus Octreotide LAR appears to be safe and efficacious in patients with advanced NETs who may have had prior exposure to chemotherapy - a finding consistent with recently conducted major trials.

Clinicopathologic profile of 470 giant cell tumors of bone from a cancer hospital in western India

Over 20 years, 470 cases of giant cell tumor of bone diagnosed at a tertiary cancer hospital were analyzed. Male predominance (57%), predilection for bones around the knee joint (42%), and occurence in the 21- to 30-year-old age group (49.1%) with 6% being in the immature skeleton are well known facts. Accurate diagnosis was possible in 66% and 88% of cases on radiology and biopsy, respectively. Tumors measured 6 to 20 cm and, in 402 cases, showed "usual" histology comprising uniformly scattered multinucleate giant cells amidst mononuclear stromal cells, together imparting a syncitium-like appearance. Presence of osteoid, hemorrhage, and aneurysmal bone cyst-like areas; spindle cells in sheets (devoid of giant cells); or storiform pattern and intravascular osteoclasts were less common. The less common histologic features posed diagnostic difficulty in the setting of a small biopsy. Treatment included intralesional curettage (33.19%), marginal excision (4.2%), wide excision (31%), or radical surgeries (14.25%). Recurrences seen in 170 cases were multiple in 47 cases. Metastases largely to the lung were recorded in 24 cases. The histology of all the tumors, namely, primary, recurrent, or metastatic was identical. Statistical analysis using the computer software SPSS (SPSS Inc, Chicago, Ill)was performed with particular reference to the unusual histologic features vs recurrence and metastasis by chi(2) test. The only statistically significant factors were occurrence in the axial skeleton vs appendicular skeleton (P = .001) and primary treatment elsewhere vs at this hospital (P = .045), each of these being associated with increased frequency for local recurrence but not metastasis.

Intraperitoneal Chemotherapy as First-Line Treatment in the Management of Epithelial Ovarian Cancer

Recent evidence has suggested improved outcomes following incorporation of intraperitoneal chemotherapy administration with intravenous systemic chemotherapy as first-line treatment of small volume residual epithelial ovarian cancer. This review focuses on the mechanism of actions of the chemotherapeutic drugs and reviews the possible reasons for the superior outcomes of intraperitoneal chemotherapy.

Genome scan of schizophrenia families in a large Veterans Affairs Cooperative Study sample: evidence for linkage to 18p11.32 and for racial heterogeneity on chromosomes 6 and 14

Genome-wide linkage analyses of schizophrenia have identified several regions that may harbor schizophrenia susceptibility genes but, given the complex etiology of the disorder, it is unlikely that all susceptibility regions have been detected. We report results from a genome scan of 166 schizophrenia families collected through the Department of Veterans Affairs Cooperative Studies Program. Our definition of affection status included schizophrenia and schizoaffective disorder, depressed type and we defined families as European American (EA) and African American (AA) based on the probands' and parents' races based on data collected by interviewing the probands. We also assessed evidence for racial heterogeneity in the regions most suggestive of linkage. The maximum LOD score across the genome was 2.96 for chromosome 18, at 0.5 cM in the combined race sample. Both racial groups showed LOD scores greater than 1.0 for chromosome 18. The empirical P-value associated with that LOD score is 0.04 assuming a single genome scan for the combined sample with race narrowly defined, and 0.06 for the combined sample allowing for broad and narrow definitions of race. The empirical P-value of observing a LOD score as large as 2.96 in the combined sample, and of at least 1.0 in each racial group, allowing for narrow and broad racial definitions, is 0.04. Evidence for the second and third largest linkage signals come solely from the AA sample on chromosomes 6 (LOD = 2.11 at 33.2 cM) and 14 (LOD = 2.13 at 51.0). The linkage evidence differed between the AA and EA samples (chromosome 6 P-value = 0.007 and chromosome 14 P-value = 0.004).

Modest evidence for linkage and possible confirmation of association between NOTCH4 and schizophrenia in a large Veterans Affairs Cooperative Study sample

Wei and Hemmings [2000: Nat Genet 25:376-377], using 80 British parent-offspring trios, identified a number of NOTCH4 variants and haplotypes that showed statistically significant evidence of association to schizophrenia. Specifically, the 10 repeat allele of a (CTG)(n) marker and the 8 repeat allele of a (TAA)(n) marker demonstrated excess transmission to affected individuals; SNP21 and haplotypes SNP2-(CTG)(n) and SNP12-SNP2-(CTG)(n) also showed significant associations. In an attempt to replicate these findings, we tested for linkage and association between the same five markers used by Wei and Hemmings in 166 families collected from a multi-center study conducted by the Department of Veterans Affairs (DVA) Cooperative Study Program (CSP). The families include 392 affected subjects (schizophrenia or schizoaffective disorder, depressed) and 216 affected sibling pairs. The families represent a mix of European Americans (n = 62, 37%), African Americans (n = 60, 36%), and racially mixed or other races (n = 44, 27%). We identified moderate evidence for linkage in the pooled race sample (LOD = 1.25) and found excess transmission of the 8 (P = 0.06) and 13 (P = 0.04) repeat alleles of the (TAA)(n) marker to African American schizophrenic subjects. The 8 and 13 repeat alleles were previously identified to be positively associated with schizophrenia by Wei and Hemmings [2000: Nat Genet 25:376-377] and Sklar et al. [2001: Nat Genet 28:126-128], respectively.

Enhanced sequential expression of G1/S cyclins during experimental epatocarcinogenesis and tyrosine phosphorylation

It is now widely accepted that cancer development is a multistage process, starting from the original cell population and ending with a malignant tumor. However, the mechanisms involved in the progressive growth of cells from normalcy to preneoplasia, and from preneoplasia to malignancy are not clear. Because tyrosine phosphorylation and dephosphorylation reactions are known to play critical roles during normal and abnormal cellular growth, we have studied the tyrosine phosphorylation, tyrosine phosphorylated proteins, and protein phosphatases during the sequential development of liver cancer. The present investigation indicated that enhanced tyrosine phosphorylation and tyrosine phosphorylated proteins, with no change in the levels of tyrosine protein phosphatases may contribute to abnormal cellular proliferation during liver carcinogenesis. We have also seen an increase in the expression of proliferating cell nuclear antigen and G1/S cyclins during tumor development.

Examination of genetic linkage of chromosome 15 to schizophrenia in a large Veterans Affairs Cooperative Study sample

Previous studies have reported genetic linkage evidence for a schizophrenia gene on chromosome 15q. Here, chromosome 15 was examined by genetic linkage analysis using 166 schizophrenia families, each with two or more affected subjects. The families, assembled from multiple centers by the Department of Veterans Affairs Cooperative Study Program, consisted of 392 sampled affected subjects and 216 affected sibling pairs. By DSM-III-R criteria, 360 subjects (91.8%) had a diagnosis of schizophrenia and 32 (8.2%) were classified as schizo-affective disorder, depressed. Participating families had diverse ethnic backgrounds. The largest single group were northern European American families (n = 62, 37%), but a substantial proportion was African American kindreds (n = 60, 36%). The chromosome 15 markers tested were spaced at intervals of approximately 10 cM over the entire chromosome and 2-5 cM for the region surrounding the alpha-7 nicotinic cholinergic receptor subunit gene (CHRNA7). These markers were genotyped and the data analyzed using semiparametric affecteds-only linkage analysis. In the European American families, there was a maximum Z-score of 1.65 between markers D15S165 and D15S1010. These markers are within 1 cM from CHRNA-7, the site previously implicated in schizophrenia. However, there was no evidence for linkage to this region in the African America kindreds.

Effect of metanil yellow and malachite green on DNA synthesis in N-nitrosodiethylamine induced preneoplastic rat livers

Metanil yellow (MY) and malachite green (MG) are textile dyes, which, despite the ban occurs unsrupulously as food colouring agents. Accordingly they constitute a serious public health hazard and are of sufficient environmental concern. We have earlier reported that both MY and MG have tumor enhancing effects on the development of hepatic preneoplastic lesions induced by N-nitrosodiethylamine in rats. In order to understand the possible mechanisms by which MY and MG enhance tumor development, in this study we have tested the effects of MY and MG on DNA synthesis and PCNA expression in preneoplastic hepatic lesions during N-nitrosodiethylamine (DEN) induced hepatocarcinogenesis in male Wistar (WR) rats. Rats were administered 200 ppm DEN through drinking water for a period of one month. Administration of DEN for a period of one month showed an upregulation of cell cycle regulatory proteins namely cyclin D1, CDK4, cyclin E and CDK2. Accordingly, in other experiments, the animals were further administered MY and MG for a period of one month following one month DEN treatment. The effects of MY and MG were monitored on the basis of cell proliferation markers--DNA synthesis and PCNA expression both by immunohistochemical and immunoblotting. Following DEN administration, MY, MG and PB showed stimulation of DNA synthesis and increased PCNA expression when compared with either the corresponding controls or only DEN treated animals. In the present study, enhancing effect of MY, MG and PB on the cell proliferation markers during DEN-induced hepatic preneoplasia in rats was observed.

Overexpression of G1/S cyclins and PCNA and their relationship to tyrosine phosphorylation and dephosphorylation during tumor promotion by metanil yellow and malachite green

Metanil yellow (MY) and Malachite green (MG) are textile dyes, which, despite the ban, occur unscrupulously as food colouring agents. Accordingly they constitute a serious public health hazard and are of sufficient environmental concern. We have earlier reported that both MY and MG have tumor promoting effects on the development of hepatic preneoplastic lesions induced by N-nitrosodiethylamine in rats. In order to understand the possible mechanism(s) by which metanil yellow (MY) and malachite green (MG) promotes liver tumor development, we have studied the tyrosine phosphorylation and protein phosphatases during tumor promotion. We have also investigated the possible overexpression of G1/S cyclins and PCNA during tumor promotion by MY and MG. The present investigation indicates that enhanced tyrosine phosphorylation is associated with no change in levels of tyrosine protein phosphatases. We have also observed an increase in the expression of PCNA and G1/S cyclins during tumor promotion. These factors collectively may contribute to the abnormal cell proliferation during tumor promotion by MY and MG.

Department of Veterans Affairs Cooperative Studies Program genetic linkage study of schizophrenia: ascertainment methods and sample description

To help clarify the genetics of schizophrenia, the Department of Veterans Affairs Cooperative Studies Program has completed data collection for a genetic linkage study of schizophrenia. This article describes the methodological details of the data collection. Subsequent articles will describe the results of our genome scan, which is now in progress. The data collection protocol included the Diagnostic Interview for Genetic Studies, the Family Interview for Genetic Studies, a review of medical records, and the collection of blood for transformation into lymphoblast cell lines. Among relatives of schizophrenic probands, we assessed auditory attention and verbal memory with neuropsychological tests. Among the 166 families ascertained for the study, 143 had a single affected sib-pair, 17 had three affected siblings, one had five affected siblings and five had two sets of affected siblings. There was a total of 216 affected sib-pairs in these families. Using the n-1 rule, these families contain 188 independent affected sib-pairs.

Effect of risperidone on QT interval and QT dispersion in the elderly

Risperidone, a relatively new antipsychotic medication, is widely used in elderly adults and is reported to have fewer side effects than existing antipsychotic drugs. Recent studies, however, have reported prolongation of QT interval and corrected QT interval, both with the prescribed dose of risperidone and with overdose. Because QT dispersion measured on the surface electrocardiogram (ECG) recently has been shown to be a marker of future arrhythmic events and to predict mortality in elderly patients in a variety of clinical situations, this study was conducted to investigate the effect of risperidone on QT dispersion and corrected QT dispersion in a group of elderly patients. These patients also were evaluated for occurrence of sudden death and/or symptoms suggestive of ventricular arrhythmia. Although risperidone prolonged QT interval, it had no significant effect on QT dispersion. Further, during the follow-up period there were no incidences of sudden death or symptoms suggestive of ventricular arrhythmia. These results indicate that risperidone can be used safely in elderly patients, who are often taking several medications, without risk of increased QT dispersion.