Systematic comparison of 3D Deep learning and classical machine learning explanations for Alzheimer's Disease detection

Black-box deep learning (DL) models trained for the early detection of Alzheimer's Disease (AD) often lack systematic model interpretation. This work computes the activated brain regions during DL and compares those with classical Machine Learning (ML) explanations. The architectures used for DL were 3D DenseNets, EfficientNets, and Squeeze-and-Excitation (SE) networks. The classical models include Random Forests (RFs), Support Vector Machines (SVMs), eXtreme Gradient Boosting (XGBoost), Light Gradient Boosting (LightGBM), Decision Trees (DTs), and Logistic Regression (LR). For explanations, SHapley Additive exPlanations (SHAP) values, Local Interpretable Model-agnostic Explanations (LIME), Gradient-weighted Class Activation Mapping (GradCAM), GradCAM++ and permutation-based feature importance were implemented. During interpretation, correlated features were consolidated into aspects. All models were trained on the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset. The validation includes internal and external validation on the Australian Imaging and Lifestyle flagship study of Ageing (AIBL) and the Open Access Series of Imaging Studies (OASIS). DL and ML models reached similar classification performances. Regarding the brain regions, both types focus on different regions. The ML models focus on the inferior and middle temporal gyri, and the hippocampus, and amygdala regions previously associated with AD. The DL models focus on a wider range of regions including the optical chiasm, the entorhinal cortices, the left and right vessels, and the 4th ventricle which were partially associated with AD. One explanation for the differences is the input features (textures vs. volumes). Both types show reasonable similarity to a ground truth Voxel-Based Morphometry (VBM) analysis. Slightly higher similarities were measured for ML models.

Data analysis with Shapley values for automatic subject selection in Alzheimer's disease data sets using interpretable machine learning

BACKGROUND

For the recruitment and monitoring of subjects for therapy studies, it is important to predict whether mild cognitive impaired (MCI) subjects will prospectively develop Alzheimer's disease (AD). Machine learning (ML) is suitable to improve early AD prediction. The etiology of AD is heterogeneous, which leads to high variability in disease patterns. Further variability originates from multicentric study designs, varying acquisition protocols, and errors in the preprocessing of magnetic resonance imaging (MRI) scans. The high variability makes the differentiation between signal and noise difficult and may lead to overfitting. This article examines whether an automatic and fair data valuation method based on Shapley values can identify the most informative subjects to improve ML classification.

METHODS

An ML workflow was developed and trained for a subset of the Alzheimer's Disease Neuroimaging Initiative (ADNI) cohort. The validation was executed for an independent ADNI test set and for the Australian Imaging, Biomarker and Lifestyle Flagship Study of Ageing (AIBL) cohort. The workflow included volumetric MRI feature extraction, feature selection, sample selection using Data Shapley, random forest (RF), and eXtreme Gradient Boosting (XGBoost) for model training as well as Kernel SHapley Additive exPlanations (SHAP) values for model interpretation.

RESULTS

The RF models, which excluded 134 of the 467 training subjects based on their RF Data Shapley values, outperformed the base models that reached a mean accuracy of 62.64% by 5.76% (3.61 percentage points) for the independent ADNI test set. The XGBoost base models reached a mean accuracy of 60.00% for the AIBL data set. The exclusion of those 133 subjects with the smallest RF Data Shapley values could improve the classification accuracy by 2.98% (1.79 percentage points). The cutoff values were calculated using an independent validation set.

CONCLUSION

The Data Shapley method was able to improve the mean accuracies for the test sets. The most informative subjects were associated with the number of ApolipoproteinE ε4 (ApoE ε4) alleles, cognitive test results, and volumetric MRI measurements.

Sociodemographic data and APOE-ε4 augmentation for MRI-based detection of amnestic mild cognitive impairment using deep learning systems

Detection and diagnosis of early and subclinical stages of Alzheimer's Disease (AD) play an essential role in the implementation of intervention and prevention strategies. Neuroimaging techniques predominantly provide insight into anatomic structure changes associated with AD. Deep learning methods have been extensively applied towards creating and evaluating models capable of differentiating between cognitively unimpaired, patients with Mild Cognitive Impairment (MCI) and AD dementia. Several published approaches apply information fusion techniques, providing ways of combining several input sources in the medical domain, which contributes to knowledge of broader and enriched quality. The aim of this paper is to fuse sociodemographic data such as age, marital status, education and gender, and genetic data (presence of an apolipoprotein E (APOE)-ε4 allele) with Magnetic Resonance Imaging (MRI) scans. This enables enriched multi-modal features, that adequately represent the MRI scan visually and is adopted for creating and modeling classification systems capable of detecting amnestic MCI (aMCI). To fully utilize the potential of deep convolutional neural networks, two extra color layers denoting contrast intensified and blurred image adaptations are virtually augmented to each MRI scan, completing the Red-Green-Blue (RGB) color channels. Deep convolutional activation features (DeCAF) are extracted from the average pooling layer of the deep learning system Inception_v3. These features from the fused MRI scans are used as visual representation for the Long Short-Term Memory (LSTM) based Recurrent Neural Network (RNN) classification model. The proposed approach is evaluated on a sub-study containing 120 participants (aMCI = 61 and cognitively unimpaired = 59) of the Heinz Nixdorf Recall (HNR) Study with a baseline model accuracy of 76%. Further evaluation was conducted on the ADNI Phase 1 dataset with 624 participants (aMCI = 397 and cognitively unimpaired = 227) with a baseline model accuracy of 66.27%. Experimental results show that the proposed approach achieves 90% accuracy and 0.90 F1-Score at classification of aMCI vs. cognitively unimpaired participants on the HNR Study dataset, and 77% accuracy and 0.83 F1-Score on the ADNI dataset.

AB1338-HPR GLOBAL PATIENT PERSPECTIVE ON TOP CHALLENGES IN LUPUS CARE AND RESEARCH PARTICIPATION

Background:

The Addressing Lupus Pillars for Health Advancement (ALPHA) Project is a global consensus initiative to identify, prioritize and address top barriers in lupus drug development, clinical care and access to care. The Lupus Foundation of America convenes ALPHA with Tufts Center for the Study of Drug Development and a Global Advisory Committee of lupus experts representing clinician-scientists, industry and patients.

Objectives:

Collect global patient input to determine alignment with the lupus clinician-scientist community on prior published consensus of top lupus barriers.

Methods:

A 23-question online Qualtrics survey was developed to identify challenges across lupus diagnosis, clinical care and research participation. The survey, available in English, Spanish, Korean and simplified Chinese, was fielded in November 2019 to people with lupus and caregivers of children <18 with lupus. SPSS 26 and SAS 9.4 were used for descriptive statistics and sub-analysis.

Results:

Analysis included only consented responses with ≥ 68% survey completion (n=3,447) received across 83 countries. 95% were female with a mean age of 45. Respondents reported being White (57%), Black or of African descent (14%), Hispanic or Latino (18%) and Asian (10%). 65% resided in the US while 35% resided in countries outside of the US. 97% were people with lupus while 3% were caregivers to children <18 with lupus.

Highest ranked challenges were similar globally and across children and adults: medication side effects, lack of treatment options and high out-of-pocket costs. Managing side effects ranked significantly higher (p<.05) outside of the US (48%) compared to US (40%). 50% of caregivers reported managing side effects as the top challenge for children compared to 43% of adults (p<.05). Research participation was low (24%) and lower among children (16%). The top reason for not participating in a clinical trial was not being asked by their doctor.

Conclusion:

This global survey revealed that medication side effects and lack of effective treatments are top challenges for people with lupus, including children. Most respondents were never asked by their doctors to participate in a clinical trial, which may explain difficulties in trial recruitment. These barriers are consistent with prior published barriers identified by the clinician-scientist community.

Acknowledgments:

ALPHA sponsors: EMD Serono, GSK, Aurinia, Mallinckrodt

Disclosure of Interests:

Karin Tse: None declared, Yaritza Peña: None declared, Kathleen Arntsen: None declared, Sang-Cheol Bae: None declared, Lauren Bloch Consultant of: Faegre Drinker Consulting is a division of Faegre Drinker Biddle & Reath, a law and consulting firm that represents patient advocacy organizations and sponsors developing drugs, Ian N. Bruce Grant/research support from: Genzyme Sanofi, GSK, and UCB, Consultant of: Eli Lilly, AstraZeneca, UCB, Iltoo, and Merck Serono, Speakers bureau: UCB, Karen Costenbader Grant/research support from: Merck, Consultant of: Astra-Zeneca, Bradley Dickerson Employee of: Aurinia, Thomas Dörner Grant/research support from: Janssen, Novartis, Roche, UCB, Consultant of: Abbvie, Celgene, Eli Lilly, Roche, Janssen, EMD, Speakers bureau: Eli Lilly, Roche, Samsung, Janssen, Kenneth Getz: None declared, Amy Kao Employee of: EMD Serono, Susan Manzi: None declared, Eric F. Morand Grant/research support from: AstraZeneca, Consultant of: AstraZeneca, Speakers bureau: AstraZeneca, Sandra Raymond: None declared, Brad H Rovin Grant/research support from: GSK, Consultant of: GSK, Laura Schanberg Grant/research support from: Sobi, BMS, Consultant of: Aurinia, UCB, Sanofi, Victoria Werth Grant/research support from: Biogen, Celgene, Gilead, Janssen, Viela, Consultant of: Biogen, Gilead, Janssen, Abbvie, GSK, Resolve, AstraZeneca, Amgen, Eli Lilly, EMD Serono, BMS, Viela, Kyowa Kirin, Joan Von Feldt Shareholder of: GSK, Employee of: GSK, David Zook Consultant of: Faegre Drinker Consulting is a division of Faegre Drinker Biddle & Reath, a law and consulting firm that represents patient advocacy organizations and sponsors developing drugs, Leslie Hanrahan: None declared

Classification of Alzheimer's Disease using volumetric features of multiple MRI scans

Volumetric measurements from magnetic resonance imaging (MRI) scans can be used to predict the future conversion to Alzheimer's disease (AD) for patients with mild cognitive impairment (MCI). Previous studies achieved good classification results using the volumes of a single as well as multiple scans per subject. The purpose of this study is to evaluate, if and how volumetric features of a baseline (BL) and a follow-up (FU) MRI scan can be combined to improve classification accuracy. For this reason, random forest (RF) models were trained on different volumetric feature sets of 513 subjects from the Alzheimer's Disease Neuroimaging Initiative (ADNI) and 22 subjects from the Australian Imaging, Biomarker & Lifestyle Flagship Study of Ageing (AIBL) database. The results show that models, which use combinations of both acquisition times yield better accuracies in comparison to the models solely based on FU or BL data. Furthermore, a clear pattern of which combination of representations performs best could not be found. The best model achieves a test classification accuracy of 75.49% (specificity: 80.52%, sensitivity: 60%). Models trained with cognitive test results and MRI data outperform models which use only MRI data. The observed results could not be reproduced on the AIBL dataset.

Brain metastases in renal cell carcinoma: management with gamma knife radiosurgery

PURPOSE

The purpose of this study was to evaluate survival and local control of brain metastases in patients with renal cell carcinoma.

METHODS AND MATERIALS

From November 1993 through March 1999, 38 radiosurgical treatments using the Leksell gamma knife unit were performed on 22 patients with renal cell carcinoma. The indications for treatment were failure after whole-brain radiation therapy or de novo treatment. All radiosurgical treatments were given on an outpatient basis. The workup included computed tomography and magnetic resonance imaging. The age of the patients ranged from 38 to 80 years (median age, 60 years). The mean minimum tumor dose was 18 Gy, and the mean volume was 3.9 cc. Previous whole-brain radiation therapy was used in 11/22 (50%) patients. Four of 22 patients presented with single metastasis. Thirteen patients were treated once, one patient was treated four times and one patient seven times for new lesions. The number of lesions treated ranged from one to 21.

RESULTS

One patient is al ive at 63 months of fol low-up. Twenty-one patients died, with a median survival of 8 months (range, 1-38 months). Eighteen of 21 patients died of nonneurologic causes. Overall local control was 98.5%. One patient developed radiation necrosis. CONCLUSIONS The long-term survival achieved in patients with renal cell carcinoma requires aggressive management, even in the presence of multiple brain metastases. Gamma knife radiosurgery for renal cell carcinoma is an effective noninvasive modality of treatment. It offers high local control rate and improved quality of life and survival.

Gamma knife radiosurgery in the treatment of patients with single and multiple brain metastases from carcinoma of the breast

PURPOSE

The purpose of the study was to evaluate the results of gamma knife radiosurgery for treatment of brain metastases from carcinoma of the breast.

MATERIALS AND METHODS

From December 1993 to July 1998, 68 women with breast carcinoma metastatic to the brain were treated with gamma knife radiosurgery at Miami Neuroscience Center in Coral Gables, Florida. The ages ranged from 25 to 80 years, with a median age of 52 years. Thirty-eight patients had previously received conventional modalities of treatment for brain metastases including whole-brain irradiation. A total of 110 treatments were given to the 68 women to an average of eight tumor sites per patient. Minimum doses ranged from 6 to 25 Gy to the 35% to 85% isodose line, with 95% of the prescribed minimum doses ranging from 15 to 24 Gy. Patients were treated for one to three lesions (n = 26), four to seven lesions (n = 18), and eight or more lesions (n = 24).

RESULTS

The median overall actuarial survival for the entire group was 7.8 months. The actuarial survival was 32% at 1 year. The median follow-up was 7.8 months. Overall local control by lesion was 94% (485/518 lesions), and average tumor volume was 3.3 cm3. Twenty-seven (40%) of 68 eligible patients survived 1 year, nine (13%) survived 2 years, and two (3%) survived more than 3 years. Fifty-one of 56 documented deaths (91%) were unrelated to brain metastases. In a subgroup of 15 patients with single brain metastases, the average tumor volume was 16.6 cm3, and local control was 73% (11/15 lesions). The 15 patients who died had a median survival of 7.7 months (range, 3 to 45.7 months).

CONCLUSIONS

Gamma knife radiosurgical treatment of patients with brain metastases from carcinoma of the breast has shifted the question of survival to that of systemic control. There was no radiation-induced dementia, and a remarkably low incidence of local failure was seen. Survival has been found to be independent of the number of lesions treated.

Enzymatic degradation of plant cell wall polysaccharides: the kinetic effect of competitive adsorption

Insoluble potato dietary fibre, isolated from potato pulp, can be enzymatically hydrolysed with the pectolytic enzyme preparation Pectinex Ultra SP from Novo Nordisk A/S, in order to produce soluble fibre. The soluble fibre has valuable functional properties for the food industry. Cloned monocomponent enzymes from Pectinex Ultra SP (arabinofuranosidase, endoglucanase II, pectin lyase, polygalacturonase I, rhamnogalacturonan acetyl esterase, rhamnogalacturonase a, rhamnogalacturonase b and xylanase I) were added in order to increase the yield. Surprisingly, however, the yield is not increased when any of the monocomponent enzymes are added. To describe the results a new model designated 'the competitive activity adsorption model' is proposed. The model is based on the fact that the enzymes are adsorbed to the substrate before action. A combination of the Langmuir adsorption isotherm and basic enzyme kinetics shows that different enzymes that adsorb competitively will have an inhibitory effect on each other and consequently decrease the hydrolysis rate and thereby the yield. The model has been confirmed by an experiment in which the fibre has been pre-treated with rhamnogalacturonan acetyl esterase.

Implantation of permanent dual chamber pacemaker in a pregnant woman by transesophageal echocardiographic guidance

A 37-year-old woman complained of fatigue and dizziness because of intermittent sinus arrest and asystole up to 5.2 seconds. She was 3 months into her pregnancy and a dual chamber permanent pacemaker was implanted by transesophageal echocardiographic guidance.

Aortic valve mass and sudden blindness

Images

[Balloon mitral valvuloplasty]

Since March 1990 we performed 100 balloon mitral valvuloplasties (BMV) in 82 females and 18 males (mean age 37 +/- 1 years; range 16 - 81) Initially we used the single shaft, dual or triple balloon system, while during the past 3 years we have been using the Inoue balloon system with step-wise inflation, monitoring with trans-thoracic echo (TTE) for immediate evaluation of mitral valve area (MVA) and/or severity of mitral regurgitation (MR), with 93% technical success. Hemodynamic data before and immediately after the procedure are in table below: [table: see text] C.O. cardiac output, DGR diastolic gradient, LAP left atrial pressure, MVA mitral valve area; all differences significant, p < 0.05. Acute complications included cardiac tamponade in 4/100; severe, acute MR in 2/100; only 5 were referred for urgent surgery. Neither cardiac tamponade nor severe MR were noted in the past 3 years. There was no periprocedural mortality in the past 5 years. Of 93 patients followed for 27 +/- 6 months (range 0-60), 90 (96%) were in NYHA classes I and II, 5 had late MVR, 2 underwent repeated BMV, and there was 1 death 16 months after the procedure. Immediate hemodynamic improvement followed BMV in most patients with pliable mitral stenosis. The Inoue system, with step-wise inflation and monitoring by TTE, proved to be a safe procedure. Symptomatic improvement continues during more than 4 years of follow-up.

[Transseptal left heart catheterization: a new application of an old invasive technique]

In this evolving era of balloon mitral valvotomy (BMV), radiofrequency ablation (RF) of left-sided bypass tracts via catheter, and hemodynamic evaluation of aortic mechanical prostheses, there has been renewed interest in transseptal left heart catheterization (TSLHC). In the 3 years 1990-1994, 122 consecutive patients were referred for TSLHC to our institute (which lacks thoracic surgical facilities). 12 patients were excluded; 10 with a LA mass proven by transesophageal echocardiography (TEE), 1 with a vascularized thrombus in the circumflex coronary system and 1 with congenital interruption of the inferior vena cava with azygous continuity. In the remaining 110 cases TSLHC was performed for interventions in 90 cases (82%) of BMV, and for left-sided catheter radiofrequency (RF) ablation in 3 (3%). For diagnostic purposes it was performed in 17 (15%) cases for hemodynamic evaluation of mechanical aortic valve prostheses. Using the Brockenbrough needle, the adult Mullins sheath system (MSS) and single plane fluoroscopy, 100% technical success was achieved. Needle puncture was not needed in 30 (27%) due to direct crossing with the MSS through a stretched foramen ovale. In 2 we had to perform SVC dye injection for better interatrial septum localization. There were no complications when TSLHC was only used for diagnostic procedures. 1 patient had perforation of the LA due to right lower pulmonary vein laceration following septal dilatation. Following stabilization by immediate pericardiocentesis, the patient was transferred for open heart surgery. There were no great vessel perforations, systemic embolization or periprocedural deaths. TSLHC can be performed quite safely with single-plane fluoroscopy without an onsight surgical team, as with an experienced staff this procedure has very low morbidity and mortality.

A Soviet nuclear pacemaker

A Soviet nuclear pacemaker is described. It is a large VOO bipolar pacemaker, probably nonprogrammable. It has unique electrode connectors. No isotope leak or excessive radiation hazard was detected. Because we had no knowledge of its end-of-life markers and because of the impossibility of assuming long-term safety, the device was removed.

[Balloon mitral valvuloplasty during pregnancy]

A 26-year-old woman with severe pliable mitral stenosis underwent successful balloon mitral valvuloplasty at the end of the second trimester of pregnancy. The indication for intervention was severe shortness of breath during most daily activities, despite combined beta-blocker and diuretic treatment (FC III, NYHA). After use of a 2 x 19 bifoil balloon there was significant clinical and hemodynamic improvement. Shortness of breath disappeared, the mean mitral valve diastolic gradient decreased from 24 to 7 mmHg and the mitral valve area increased from 0.8 to 1.8 cm. There were no complications of the treatment, and estimated radiation exposure of the fetus was less than 0.2 rad. Subsequent abdominal ultrasound examination revealed normal fetal functioning. 2.5 months after mitral dilatation delivery was normal. Balloon mitral valvuloplasty may safely be used instead of surgery as a palliative procedure for relief of symptoms in pregnant women with severe pliable mitral stenosis.

Idiopathic hypertrophic sub-aortic stenosis (IHSS): a new diagnostic method using ECG-gated thallium-201 myocardial scintigraphy

A new diagnostic method for idiopathic hypertrophic sub-aortic stenosis (IHSS), using synchronized ECG gated 201Tl myocardial scintigraphy, is described. Twenty patients previously diagnosed as IHSS were evaluated by sequential three-view ECG gated 201Tl and 45 degrees left anterior oblique multigated (MUGA) cardiac blood pool imaging (Group A). The results were compared with two control groups: 20 patients with no previous history of heart disease (Group B), and 20 patients with coronary ischaemic heart disease (Group C). We suggest the following combination of scintigraphic signs as typical and diagnostic to obstructive IHSS: Elevated left ventricular ejection fraction (mean 83.9% +/- 6.15S.D. versus 72.5% +/- 5.35S.D. and 51.2% +/- 13.65S.D. in Groups B and C respectively); Systolic left ventricular cavity obliteration (71% of obstructive IHSS patients versus 0% in Groups B and C); A ratio of more than 1.3:1 in septal to free wall thickness (100% of patients with IHSS); Perfusion/wall-motion mismatch of septum was demonstrated in all (100%) of patients with IHSS (versus 0% in Group B and 10% in Group C). We suggest this last finding as a new specific diagnostic sign for IHSS, with the ability to differentiate between patients with ischaemic coronary heart disease and IHSS patients; whilst both groups may present chest pains, the first group was characterized by 'matching' of perfusion and motion of the affected myocardial wall while the patients with IHSS had 'mismatching' of these parameters. We suggest gated 201Tl myocardial scintigraphy as an additional sensitive tool for the early diagnosis and evaluation of IHSS patients.

Combined administration of propranolol and quinidine in the conversion of paroxysmal atrial fibrillation

Quinidine was combined with propranolol to restore sinus rhythm in cases of paroxysmal atrial fibrillation (PAF) in 42 patients. The treatment was started with propranolol and quinidine at a test oral dose of 10 mg and 0.2 g, respectively, thereafter a dose of 0.4 g of quinidine and 10 mg of propranolol every two hours was added. The procedure was stopped when sinus rhythm was achieved, or when the administered quinidine reached the total dose of 1.6 g. Sinus rhythm was restored in 37 patients. The side effects were gastrointestinal disorders in 16 patients, atypical ventricular tachycardia in one case, and embolism of the left femoral artery in another patient.

The relationships among sexual beliefs, attitudes, experience, and homophobia

Male and female subjects were given a series of questionnaires to assess their attitudes, behaviors, and experiences in relation to homosexuality. The findings indicated the presence of two systems, one dealing with affective orientation and the other with general beliefs (learned problem or physiological problem) about the origins of homosexuality. It was found that the greatest dislike toward homosexuals existed in those subjects who responded with negative affect and believed that homosexuality was a learned problem. Avoidance of social situations where homosexuals are present was evidenced in subjects who responded with negative affect and believed homosexuality was due to genetic factors.

The diagnosis and early complications of right ventricular infarction

Studies of pathology have shown that involvement of the right ventricle (RV) in acute myocardial infarction (AMI) is relatively common. Our experience in the noninvasive diagnosis of RVAMI and its early prognosis is presented. Sixty patients with AMI were prospectively studied: 40 patients with inferior AMI and 20 patients with anterior AMI. The evaluation was made by: (1) CLINICAL FINDINGS: hypotension, congestive heart failure, sinus bradycardia less than 40/min, dysrhythmia, and conduction disturbances; (2) ECG record including precordial lead (V4R); (3) Radioisotope heart scintigraphy: 99mTc-PYP infarct scintigraphy and multigated acquisition MUGA blood pool scanning. Of the 40 patients diagnosed as having inferior AMI, 20 cases (50%) were found to be associated with RVAMI. All of them were diagnosed by positive radionuclide studies, and 17 (85%) also demonstrated a ST segment elevation of 0.1 mV, and pathological Q waves in the V4R lead. The ejection fraction (EF) of RV was found to be significantly decreased in patients with RVAMI compared with the other group (mean, 27% versus 57%). Among the 20 patients with RVAMI, 16 (80%) showed various complications during the hospitalization period, versus 9 patients (45%) from the group with inferior AMI. The most common complication in RVAMI patients was conduction disturbances (7 of 20 versus 2 of 20 patients). The clinical and prognostic importance of the early diagnosis of RVAMI is stressed.