Explainable Artificial Intelligence for Bias Detection in COVID CT-Scan Classifiers.

PROBLEM: An application of Explainable Artificial Intelligence Methods for COVID CT-Scan classifiers is presented. MOTIVATION: It is possible that classifiers are using spurious artifacts in dataset images to achieve high performances, and such explainable techniques can help identify this issue. AIM: For this purpose, several approaches were used in tandem, in order to create a complete overview of the classificatios. METHODOLOGY: The techniques used included GradCAM, LIME, RISE, Squaregrid, and direct Gradient approaches (Vanilla, Smooth, Integrated). MAIN RESULTS: Among the deep neural networks architectures evaluated for this image classification task, VGG16 was shown to be most affected by biases towards spurious artifacts, while DenseNet was notably more robust against them. Further impacts: Results further show that small differences in validation accuracies can cause drastic changes in explanation heatmaps for DenseNet architectures, indicating that small changes in validation accuracy may have large impacts on the biases learned by the networks. Notably, it is important to notice that the strong performance metrics achieved by all these networks (Accuracy, F1 score, AUC all in the 80 to 90% range) could give users the erroneous impression that there is no bias. However, the analysis of the explanation heatmaps highlights the bias.

Local Interpretable Model-Agnostic Explanations for Classification of Lymph Node Metastases.

An application of explainable artificial intelligence on medical data is presented. There is an increasing demand in machine learning literature for such explainable models in health-related applications. This work aims to generate explanations on how a Convolutional Neural Network (CNN) detects tumor tissue in patches extracted from histology whole slide images. This is achieved using the "locally-interpretable model-agnostic explanations" methodology. Two publicly-available convolutional neural networks trained on the Patch Camelyon Benchmark are analyzed. Three common segmentation algorithms are compared for superpixel generation, and a fourth simpler parameter-free segmentation algorithm is proposed. The main characteristics of the explanations are discussed, as well as the key patterns identified in true positive predictions. The results are compared to medical annotations and literature and suggest that the CNN predictions follow at least some aspects of human expert knowledge.

A novel vaccine based on SARS-CoV-2 CD4+ and CD8+ T cell conserved epitopes from variants Alpha to Omicron.

COVID-19 caused, as of September, 1rst, 2022, 599,825,400 confirmed cases, including 6,469,458 deaths. Currently used vaccines reduced severity and mortality but not virus transmission or reinfection by different strains. They are based on the Spike protein of the Wuhan reference virus, which although highly antigenic suffered many mutations in SARS-CoV-2 variants, escaping vaccine-generated immune responses. Multiepitope vaccines based on 100% conserved epitopes of multiple proteins of all SARS-CoV-2 variants, rather than a single highly mutating antigen, could offer more long-lasting protection. In this study, a multiepitope multivariant vaccine was designed using immunoinformatics and in silico approaches. It is composed of highly promiscuous and strong HLA binding CD4+ and CD8+ T cell epitopes of the S, M, N, E, ORF1ab, ORF 6 and ORF8 proteins. Based on the analysis of one genome per WHO clade, the epitopes were 100% conserved among the Wuhan-Hu1, Alpha, Beta, Gamma, Delta, Omicron, Mµ, Zeta, Lambda and R1 variants. An extended epitope-conservancy analysis performed using GISAID metadata of 3,630,666 SARS-CoV-2 genomes of these variants and the additional genomes of the Epsilon, Lota, Theta, Eta, Kappa and GH490 R clades, confirmed the high conservancy of the epitopes. All but one of the CD4 peptides showed a level of conservation greater than 97% among all genomes. All but one of the CD8 epitopes showed a level of conservation greater than 96% among all genomes, with the vast majority greater than 99%. A multiepitope and multivariant recombinant vaccine was designed and it was stable, mildly hydrophobic and non-toxic. The vaccine has good molecular docking with TLR4 and promoted, without adjuvant, strong B and Th1 memory immune responses and secretion of high levels of IL-2, IFN-γ, lower levels of IL-12, TGF-ß and IL-10, and no IL-6. Experimental in vivo studies should validate the vaccine's further use as preventive tool with cross-protective properties.

Evolved explainable classifications for lymph node metastases.

A novel evolutionary approach for Explainable Artificial Intelligence is presented: the "Evolved Explanations" model (EvEx). This methodology combines Local Interpretable Model Agnostic Explanations (LIME) with Multi-Objective Genetic Algorithms to allow for automated segmentation parameter tuning in image classification tasks. In this case, the dataset studied is Patch-Camelyon, comprised of patches from pathology whole slide images. A publicly available Convolutional Neural Network (CNN) was trained on this dataset to provide a binary classification for presence/absence of lymph node metastatic tissue. In turn, the classifications are explained by means of evolving segmentations, seeking to optimize three evaluation goals simultaneously. The final explanation is computed as the mean of all explanations generated by Pareto front individuals, evolved by the developed genetic algorithm. To enhance reproducibility and traceability of the explanations, each of them was generated from several different seeds, randomly chosen. The observed results show remarkable agreement between different seeds. Despite the stochastic nature of LIME explanations, regions of high explanation weights proved to have good agreement in the heat maps, as computed by pixel-wise relative standard deviations. The found heat maps coincide with expert medical segmentations, which demonstrates that this methodology can find high quality explanations (according to the evaluation metrics), with the novel advantage of automated parameter fine tuning. These results give additional insight into the inner workings of neural network black box decision making for medical data.

Convolutional ensembles for Arabic Handwritten Character and Digit Recognition.

A learning algorithm is proposed for the task of Arabic Handwritten Character and Digit recognition. The architecture consists on an ensemble of different Convolutional Neural Networks. The proposed training algorithm uses a combination of adaptive gradient descent on the first epochs and regular stochastic gradient descent in the last epochs, to facilitate convergence. Different validation strategies are tested, namely Monte Carlo Cross-Validation and K-fold Cross Validation. Hyper-parameter tuning was done by using the MADbase digits dataset. State of the art validation and testing classification accuracies were achieved, with average values of 99.74% and 99.47% respectively. The same algorithm was then trained and tested with the AHCD character dataset, also yielding state of the art validation and testing classification accuracies: 98.60% and 98.42% respectively.

Safe exposure distances for transcranial magnetic stimulation based on computer simulations.

The results of a computer simulation examining the compliance of a given transcranial magnetic stimulation device to the 2010 International Commission on Non-Ionizing Radiation Protection (ICNIRP) guidelines are presented. The objective was to update the safe distance estimates with the most current safety guidelines, as well as comparing these to values reported in previous publications. The 3D data generated was compared against results available in the literature, regarding the MCB-70 coil by Medtronic. Regarding occupational exposure, safe distances of 1.46 m and 0.96 m are derived from the simulation according to the 2003 and 2010 ICNIRP guidelines, respectively. These values are then compared to safe distances previously reported in other studies.

Leishmania donovani Nucleoside Hydrolase (NH36) Domains Induce T-Cell Cytokine Responses in Human Visceral Leishmaniasis.

Development of immunoprotection against visceral leishmaniasis (VL) focused on the identification of antigens capable of inducing a Th1 immune response. Alternatively, antigens targeting the CD8 and T-regulatory responses are also relevant in VL pathogenesis and worthy of being included in a preventive human vaccine. We assessed in active and cured patients and VL asymptomatic subjects the clinical signs and cytokine responses to the Leishmania donovani nucleoside hydrolase NH36 antigen and its N-(F1), central (F2) and C-terminal (F3) domains. As markers of VL resistance, the F2 induced the highest levels of IFN-γ, IL-1ß, and TNF-α and, together with F1, the strongest secretion of IL-17, IL-6, and IL-10 in DTH+ and cured subjects. F2 also promoted the highest frequencies of CD3+CD4+IL-2+TNF-α-IFN-γ-, CD3+CD4+IL-2+TNF-α+IFN-γ-, CD3+CD4+IL-2+TNF-α-IFN-γ+, and CD3+CD4+IL-2+TNF-α+IFN-γ+ T cells in cured and asymptomatic subjects. Consistent with this, the IFN-γ increase was correlated with decreased spleen (R = -0.428, P = 0.05) and liver sizes (R = -0.428, P = 0.05) and with increased hematocrit counts (R = 0.532, P = 0.015) in response to F1 domain, and with increased hematocrit (R = 0.512, P 0.02) and hemoglobin counts (R = 0.434, P = 0.05) in response to F2. Additionally, IL-17 increases were associated with decreased spleen and liver sizes in response to F1 (R = -0.595, P = 0.005) and F2 (R = -0.462, P = 0.04). Conversely, F1 and F3 increased the CD3+CD8+IL-2+TNF-α-IFN-γ-, CD3+CD8+IL-2+TNF-α+IFN-γ-, and CD3+CD8+IL-2+TNF-α+IFN-γ+ T cell frequencies of VL patients correlated with increased spleen and liver sizes and decreased hemoglobin and hematocrit values. Therefore, cure and acquired resistance to VL correlate with the CD4+-Th1 and Th-17 T-cell responses to F2 and F1 domains. Clinical VL outcomes, by contrast, correlate with CD8+ T-cell responses against F3 and F1, potentially involved in control of the early infection. The in silico-predicted NH36 epitopes are conserved and bind to many HL-DR and HLA and B allotypes. No human vaccine against Leishmania is available thus far. In this investigation, we identified the NH36 domains and epitopes that induce CD4+ and CD8+ T cell responses, which could be used to potentiate a human universal T-epitope vaccine against leishmaniasis.

Leishmania donovani Nucleoside Hydrolase Terminal Domains in Cross-Protective Immunotherapy Against Leishmania amazonensis Murine Infection.

Nucleoside hydrolases of the Leishmania genus are vital enzymes for the replication of the DNA and conserved phylogenetic markers of the parasites. Leishmania donovani nucleoside hydrolase (NH36) induced a main CD4(+) T cell driven protective response against L. chagasi infection in mice which is directed against its C-terminal domain. In this study, we used the three recombinant domains of NH36: N-terminal domain (F1, amino acids 1-103), central domain (F2 aminoacids 104-198), and C-terminal domain (F3 amino acids 199-314) in combination with saponin and assayed their immunotherapeutic effect on Balb/c mice previously infected with L. amazonensis. We identified that the F1 and F3 peptides determined strong cross-immunotherapeutic effects, reducing the size of footpad lesions to 48 and 64%, and the parasite load in footpads to 82.6 and 81%, respectively. The F3 peptide induced the strongest anti-NH36 antibody response and intradermal response (IDR) against L. amazonenis and a high secretion of IFN-γ and TNF-α with reduced levels of IL-10. The F1 vaccine, induced similar increases of IgG2b antibodies and IFN-γ and TNF-α levels, but no IDR and no reduction of IL-10. The multiparameter flow cytometry analysis was used to assess the immune response after immunotherapy and disclosed that the degree of the immunotherapeutic effect is predicted by the frequencies of the CD4(+) and CD8(+) T cells producing IL-2 or TNF-α or both. Total frequencies and frequencies of double-cytokine CD4 T cell producers were enhanced by F1 and F3 vaccines. Collectively, our multifunctional analysis disclosed that immunotherapeutic protection improved as the CD4 responses progressed from 1+ to 2+, in the case of the F1 and F3 vaccines, and as the CD8 responses changed qualitatively from 1+ to 3+, mainly in the case of the F1 vaccine, providing new correlates of immunotherapeutic protection against cutaneous leishmaniasis in mice based on T-helper TH1 and CD8(+) mediated immune responses.