Medical Image Classification Using Light-Weight CNN With Spiking Cortical Model Based Attention Module

IEEE Journal of Biomedical and Health Informatics, Год журнала: 2023, Номер 27(4), С. 1991 - 2002

Опубликована: Фев. 1, 2023

In the field of disease diagnosis where only a small dataset medical images may be accessible, light-weight convolutional neural network (CNN) has become popular because it can help to avoid over-fitting problem and improve computational efficiency. However, feature extraction capability CNN is inferior that heavy-weight counterpart. Although attention mechanism provides feasible solution this problem, existing modules, such as squeeze excitation module block module, have insufficient non-linearity, thereby influencing ability discover key features. To address issue, we proposed spiking cortical model based global local (SCM-GL) module. The SCM-GL analyzes input maps in parallel decomposes each map into several components according relation between pixels their neighbors. are weighted summed obtain mask. Besides, mask produced by discovering correlation distant map. final generated combining masks, multiplied original so important highlighted facilitate accurate diagnosis. appreciate performance some mainstream modules been embedded models for comparison. Experiments on classification brain MR, chest X-ray, osteosarcoma image datasets demonstrate significantly evaluated enhancing suspected lesions generally superior state-of-the-art terms accuracy, recall, specificity F1 score.

Язык: Английский

---

Explainable Machine-Learning Models for COVID-19 Prognosis Prediction Using Clinical, Laboratory and Radiomic Features

IEEE Access, Год журнала: 2023, Номер 11, С. 121492 - 121510

Опубликована: Янв. 1, 2023

The SARS-CoV-2 virus pandemic had devastating effects on various aspects of life: clinical cases, ranging from mild to severe, can lead lung failure and death. Due the high incidence, data-driven models support physicians in patient management. explainability interpretability machine-learning are mandatory scenarios. In this work, clinical, laboratory radiomic features were used train for COVID-19 prognosis prediction. Using Explainable AI algorithms, a multi-level explainable method was proposed taking into account developer involved stakeholder (physician, patient) perspectives. A total 1023 extracted 1589 Chest X-Ray images (CXR), combined with 38 clinical/laboratory features. After pre-processing selection phases, 40 CXR 23 Support Vector Machine Random Forest classifiers exploring three feature strategies. combination both radiomic, enabled higher performance resulting models. intelligibility allowed us validate models' findings. According medical literature, LDH, PaO2 CRP most predictive Instead, ZoneEntropy HighGrayLevelZoneEmphasis - indicative heterogeneity/uniformity texture discriminating Our best model, exploiting classifier signature composed features, achieved AUC=0.819, accuracy=0.733, specificity=0.705, sensitivity=0.761 test set. including explainability, allows make strong assumptions, confirmed by literature insights.

Язык: Английский

---

Improving deep neural network generalization and robustness to background bias via layer-wise relevance propagation optimization

Nature Communications, Год журнала: 2024, Номер 15(1)

Опубликована: Янв. 4, 2024

Abstract Features in images’ backgrounds can spuriously correlate with the classes, representing background bias. They influence classifier’s decisions, causing shortcut learning (Clever Hans effect). The phenomenon generates deep neural networks (DNNs) that perform well on standard evaluation datasets but generalize poorly to real-world data. Layer-wise Relevance Propagation (LRP) explains DNNs’ decisions. Here, we show optimization of LRP heatmaps minimize bias classifiers, hindering learning. By not increasing run-time computational cost, approach is light and fast. Furthermore, it applies virtually any classification architecture. After injecting synthetic backgrounds, compared our (dubbed ISNet) eight state-of-the-art DNNs, quantitatively demonstrating its superior robustness Mixed are common for COVID-19 tuberculosis chest X-rays, fostering focusing lungs, ISNet reduced Thus, generalization performance external (out-of-distribution) test databases significantly surpassed all implemented benchmark models.

Язык: Английский

---

Synergizing AI, IoT, and Blockchain for Diagnosing Pandemic Diseases in Smart Cities: Challenges and Opportunities

Sustainable Machine Intelligence Journal, Год журнала: 2024, Номер 7

Опубликована: Май 25, 2024

The advent of smart cities has paved the way for transformative advancements in healthcare, particularly domain disease diagnosis. In wake COVID-19 pandemic, accurate and timely identification Pandemic diseases become paramount. This paper explores challenges opportunities synergizing Artificial Intelligence (AI), Internet Things (IoT), Blockchain technologies diagnosis cities. study provides an overview each technology its relevance to sustainable healthcare cities, emphasizing potential analyzing medical data making informed decisions. We also explore how IoT devices can contribute surveillance, enabling real-time collection remote healthcare. Additionally, we discuss ensuring secure transparent systems. Following, synergistic integrating AI, IoT, blockchain, their combined strengths enhance accuracy, efficiency, security systems Moreover, highlights these research implementation, underlining significance findings demonstrate that convergence blockchain speed accuracy diagnosing diseases, leading more effective containment management strategies.

Язык: Английский

---

A comprehensive review on transformer network for natural and medical image analysis

Computer Science Review, Год журнала: 2024, Номер 53, С. 100648 - 100648

Опубликована: Июнь 14, 2024

Язык: Английский

---

Multimodal explainability via latent shift applied to COVID-19 stratification

Pattern Recognition, Год журнала: 2024, Номер 156, С. 110825 - 110825

Опубликована: Июль 24, 2024

We are witnessing a widespread adoption of artificial intelligence in healthcare. However, most the advancements deep learning this area consider only unimodal data, neglecting other modalities. Their multimodal interpretation necessary for supporting diagnosis, prognosis and treatment decisions. In work we present architecture, which jointly learns modality reconstructions sample classifications using tabular imaging data. The explanation decision taken is computed by applying latent shift that, simulates counterfactual prediction revealing features each that contribute to quantitative score indicating importance. validate our approach context COVID-19 pandemic AIforCOVID dataset, contains data early identification patients at risk severe outcome. results show proposed method provides meaningful explanations without degrading classification performance.

Язык: Английский

---

Comprehensive Survey of Using Machine Learning in the COVID-19 Pandemic

Diagnostics, Год журнала: 2021, Номер 11(7), С. 1155 - 1155

Опубликована: Июнь 24, 2021

Since December 2019, the global health population has faced rapid spreading of coronavirus disease (COVID-19). With incremental acceleration number infected cases, World Health Organization (WHO) reported COVID-19 as an epidemic that puts a heavy burden on healthcare sectors in almost every country. The potential artificial intelligence (AI) this context is difficult to ignore. AI companies have been racing develop innovative tools contribute arm world against pandemic and minimize disruption it may cause. main objective study survey decisive role technology used fight pandemic. Five significant applications for were found, including (1) diagnosis using various data types (e.g., images, sound, text); (2) estimation possible future spread based current confirmed cases; (3) association between infection patient characteristics; (4) vaccine development drug interaction; (5) supporting applications. This also introduces comparison datasets. Based limitations literature, review highlights open research challenges could inspire application COVID-19.

Язык: Английский

---

Public Covid-19 X-ray datasets and their impact on model bias – A systematic review of a significant problem

Medical Image Analysis, Год журнала: 2021, Номер 74, С. 102225 - 102225

Опубликована: Сен. 27, 2021

Computer-aided-diagnosis and stratification of COVID-19 based on chest X-ray suffers from weak bias assessment limited quality-control. Undetected induced by inappropriate use datasets, improper consideration confounders prevents the translation prediction models into clinical practice. By adopting established tools for model evaluation to task evaluating this study provides a systematic appraisal publicly available determining their potential sources bias. Only 9 out more than hundred identified datasets met at least criteria proper risk could be analysed in detail. Remarkably most utilised 201 papers published peer-reviewed journals, are not among these thus leading with high This raises concerns about suitability such use. review highlights description employed modelling aids researchers select suitable task.

Язык: Английский

---

Role of Artificial Intelligence in COVID-19 Detection

Sensors, Год журнала: 2021, Номер 21(23), С. 8045 - 8045

Опубликована: Дек. 1, 2021

The global pandemic of coronavirus disease (COVID-19) has caused millions deaths and affected the livelihood many more people. Early rapid detection COVID-19 is a challenging task for medical community, but it also crucial in stopping spread SARS-CoV-2 virus. Prior substantiation artificial intelligence (AI) various fields science encouraged researchers to further address this problem. Various imaging modalities including X-ray, computed tomography (CT) ultrasound (US) using AI techniques have greatly helped curb outbreak by assisting with early diagnosis. We carried out systematic review on state-of-the-art applied CT, US images detect COVID-19. In paper, we discuss approaches used authors significance these research efforts, potential challenges, future trends related implementation an system during pandemic.

Язык: Английский

---

Checklist for responsible deep learning modeling of medical images based on COVID-19 detection studies

Pattern Recognition, Год журнала: 2021, Номер 118, С. 108035 - 108035

Опубликована: Май 21, 2021

The sudden outbreak and uncontrolled spread of COVID-19 disease is one the most important global problems today. In a short period time, it has led to development many deep neural network models for detection with modules explainability. this work, we carry out systematic analysis various aspects proposed models. Our revealed numerous mistakes made at different stages data acquisition, model development, explanation construction. overview approaches in surveyed Machine Learning articles indicate typical errors emerging from lack understanding radiography domain. We present perspective both: experts field - radiologists learning engineers dealing explanations. final result checklist minimum conditions be met by reliable diagnostic model.

Язык: Английский

---