Transfer Learning for the Detection and Diagnosis of Types of Pneumonia including Pneumonia Induced by COVID-19 from Chest X-ray Images

Saliency-driven explainable deep learning in medical imaging: bridging visual explainability and statistical quantitative analysis

Deep learning shows great promise for medical image analysis but often lacks explainability, hindering its adoption in healthcare. Attribution techniques that explain model reasoning can potentially increase trust in deep learning among clinical stakeholders. In the literature, much of the research on attribution in medical imaging focuses on visual inspection rather than statistical quantitative analysis.In this paper, we proposed an image-based saliency framework to enhance the explainability of deep learning models in medical image analysis. We use adaptive path-based gradient integration, gradient-free techniques, and class activation mapping along with its derivatives to attribute predictions from brain tumor MRI and COVID-19 chest X-ray datasets made by recent deep convolutional neural network models.The proposed framework integrates qualitative and statistical quantitative assessments, employing Accuracy Information Curves (AICs) and Softmax Information Curves (SICs) to measure the effectiveness of saliency methods in retaining critical image information and their correlation with model predictions. Visual inspections indicate that methods such as ScoreCAM, XRAI, GradCAM, and GradCAM++ consistently produce focused and clinically interpretable attribution maps. These methods highlighted possible biomarkers, exposed model biases, and offered insights into the links between input features and predictions, demonstrating their ability to elucidate model reasoning on these datasets. Empirical evaluations reveal that ScoreCAM and XRAI are particularly effective in retaining relevant image regions, as reflected in their higher AUC values. However, SICs highlight variability, with instances of random saliency masks outperforming established methods, emphasizing the need for combining visual and empirical metrics for a comprehensive evaluation.The results underscore the importance of selecting appropriate saliency methods for specific medical imaging tasks and suggest that combining qualitative and quantitative approaches can enhance the transparency, trustworthiness, and clinical adoption of deep learning models in healthcare. This study advances model explainability to increase trust in deep learning among healthcare stakeholders by revealing the rationale behind predictions. Future research should refine empirical metrics for stability and reliability, include more diverse imaging modalities, and focus on improving model explainability to support clinical decision-making.

Automatic Mushroom Species Classification Model for Foodborne Disease Prevention Based on Vision Transformer

Mushrooms are the fleshy, spore-bearing structure of certain fungi, produced by a group of mycelia and buried in a substratum. Mushrooms are classified as edible, medicinal, and poisonous. However, many poisoning incidents occur yearly by consuming wild mushrooms. Thousands of poisoning incidents are reported each year globally, and 80% of these are from unidentified species of mushrooms. Mushroom poisoning is one of the most serious food safety issues worldwide. Motivated by this problem, this study uses an open-source mushroom dataset and employs several data augmentation approaches to decrease the probability of model overfitting. We propose a novel deep learning pipeline (ViT-Mushroom) for mushroom classification using the Vision Transformer large network (ViT-L/32). We compared the performance of our method against that of a convolutional neural network (CNN). We visualized the high-dimensional outputs of the ViT-L/32 model to achieve the interpretability of ViT-L/32 using the t-distributed stochastic neighbor embedding (t-SNE) method. The results show that ViT-L/32 is the best on the testing dataset, with an accuracy score of 95.97%. These results surpass previous approaches in reducing intraclass variability and generating well-separated feature embeddings. The proposed method is a promising deep learning model capable of automatically classifying mushroom species, helping wild mushroom consumers avoid eating toxic mushrooms, safeguarding food safety, and preventing public health incidents of food poisoning. The results will offer valuable resources for food scientists, nutritionists, and the public health sector regarding the safety and quality of mushrooms.

Development of a Fast Fourier Transform-based Analytical Method for COVID-19 Diagnosis from Chest X-Ray Images Using GNU Octave

Purpose

Many artificial intelligence-based computational procedures are developed to diagnose COVID-19 infection from chest X-ray (CXR) images, as diagnosis by CXR imaging is less time consuming and economically cheap compared to other detection procedures. Due to unavailability of skilled computer professionals and high computer architectural resource, majority of the employed methods are difficult to implement in rural and poor economic settings. Majority of such reports are devoid of codes and ignores related diseases (pneumonia). The absence of codes makes limitation in applying them widely. Hence, validation testing followed by evidence-based medical practice is difficult. The present work was aimed to develop a simple method that requires a less computational expertise and minimal level of computer resource, but with statistical inference.

Materials and Methods

A Fast Fourier Transform-based (FFT) method was developed with GNU Octave, a free and open-source platform. This was employed to the images of CXR for further analysis. For statistical inference, two variables, i.e., the highest peak and number of peaks in the FFT distribution plot were considered.

Results

The comparison of mean values among different groups (normal, COVID-19, viral, and bacterial pneumonia [BP]) showed statistical significance, especially when compared to normal, except between viral and BP groups.

Conclusion

Parametric statistical inference from our result showed high level of significance (P < 0.001). This is comparable to the available artificial intelligence-based methods (where accuracy is about 94%). Developed method is easy, availability with codes, and requires a minimal level of computer resource and can be tested with a small sample size in different demography, and hence, be implemented in a poor socioeconomic setting.

Multi-Classification of Chest X-rays for COVID-19 Diagnosis Using Deep Learning Algorithms

Accurate detection of COVID-19 is of immense importance to help physicians intervene with appropriate treatments. Although RT-PCR is routinely used for COVID-19 detection, it is expensive, takes a long time, and is prone to inaccurate results. Currently, medical imaging-based detection systems have been explored as an alternative for more accurate diagnosis. In this work, we propose a multi-level diagnostic framework for the accurate detection of COVID-19 using X-ray scans based on transfer learning. The developed framework consists of three stages, beginning with a pre-processing step to remove noise effects and image resizing followed by a deep learning architecture utilizing an Xception pre-trained model for feature extraction from the pre-processed image. Our design utilizes a global average pooling (GAP) layer for avoiding over-fitting, and an activation layer is added in order to reduce the losses. Final classification is achieved using a softmax layer. The system is evaluated using different activation functions and thresholds with different optimizers. We used a benchmark dataset from the kaggle website. The proposed model has been evaluated on 7395 images that consist of 3 classes (COVID-19, normal and pneumonia). Additionally, we compared our framework with the traditional pre-trained deep learning models and with other literature studies. Our evaluation using various metrics showed that our framework achieved a high test accuracy of 99.3% with a minimum loss of 0.02 using the LeakyReLU activation function at a threshold equal to 0.1 with the RMSprop optimizer. Additionally, we achieved a sensitivity and specificity of 99 and F1-Score of 99.3% with only 10 epochs and a 10−4 learning rate.

SARS-CoV-2 Infection Prevalence in the Population of South-Eastern Poland

COVID-19 outbreak began in Wuhan, China, and has spread to other continents, including Europe, placing pressure on healthcare systems. Poland is one of the European countries with the highest number of SARS-CoV-2 infections and COVID-19-related deaths. The aim of this study was to analyze the presence of SARS-CoV-2 in the population of south-eastern Poland. The correlation between viral infection and demographic data (gender, age, place of residence) and cancer was also investigated. A total of 44,801 samples were tested, of which 4862 cases were diagnosed with SARS-CoV-2 infections. A total of 14,970 samples were tested in cancer patients. The RT-PCR method was used to detect viral nucleic acid. In this study, significantly, the highest rate of virus detection was among people living in Lublin and the lowest among people living in a small town (p < 0.0001). Moreover, there was no significant relationship between sex and the frequency of virus detection. The highest number of SARS-CoV-2 infections was observed in the age groups 10-19, 20-29, 30-39, and 90+ (p = 0.0001). In cancer patients, the percentage of positive cases was significantly lower than in the rest (p = 0.0001).