Researchers at Manipal Academy of Higher Education outline a systematic review protocol to evaluate ML-based AI diagnostic tools for tropical fevers such as dengue, malaria, scrub typhus and chikungunya. They will conduct comprehensive database searches, apply QUADAS-2 for bias assessment, extract TP/TN/FP/FN metrics, and perform meta-analysis using Meta-DiSc and HSROC modeling to pool sensitivity and specificity outcomes.
Key points
Systematic search across Medline, Embase, Cochrane and Scopus for ML-based diagnostic studies in tropical fevers.
Dual independent screening with Cohen’s kappa, QUADAS-2 bias assessment and extraction of TP/TN/FP/FN performance metrics.
Meta-analysis via Meta-DiSc and hierarchical SROC modeling to pool sensitivity, specificity and explore heterogeneity.
Why it matters:
Pooling ML-based diagnostic performance metrics for tropical fevers guides development of accurate, scalable AI tools in resource-limited settings.
Q&A
What is QUADAS-2?
Why include only cross-sectional studies?
What is HSROC modeling?
How do TP, TN, FP, FN metrics work?
What is Meta-DiSc software?
Read full article
Academy
Machine Learning in Disease Diagnosis
Introduction
Machine learning (ML) applies algorithms that learn patterns from medical data to assist clinicians in diagnosing diseases. In the context of tropical fevers—dengue, malaria, scrub typhus, leptospirosis, chikungunya, influenza and typhoid—ML models can analyze clinical signs, laboratory tests and epidemiological factors to distinguish between similar febrile illnesses.
How ML Models Work
ML models are trained on labeled datasets where each patient record includes input features (age, symptoms, blood counts, geography) and the confirmed diagnosis. Common approaches include:
- Supervised Learning: Models like logistic regression, random forests or neural networks learn to map input data to disease labels by minimizing prediction errors during training.
- Deep Learning: Convolutional neural networks (CNNs) and other architectures process complex data, such as blood smear images, directly extracting features for classification.
- Rule-Based Systems: Expert-driven sets of if-then rules that codify clinical guidelines in a deterministic fashion.
Key Performance Metrics
Diagnostic accuracy relies on four counts:
- True Positives (TP): Cases correctly identified as disease-positive.
- True Negatives (TN): Cases correctly identified as disease-negative.
- False Positives (FP): Healthy cases incorrectly flagged as positive.
- False Negatives (FN): Diseased cases missed by the model.
From these, sensitivity (TP/(TP+FN)) and specificity (TN/(TN+FP)) quantify a model’s ability to detect disease correctly and avoid false alarms.
Systematic Reviews and Meta-Analysis
When multiple studies report ML diagnostic performance, a systematic review collates all eligible cross-sectional research. Key steps include:
- Defining inclusion criteria and search terms across databases (Medline, Embase, Cochrane, Scopus).
- Dual screening of titles/abstracts and full texts with inter-rater agreement assessed by Cohen’s kappa.
- Evaluating study quality using QUADAS-2, which rates risk of bias across patient selection, index test, reference standard and flow/timing.
- Extracting TP, TN, FP, FN values for each study to construct 2×2 tables.
Pooling Data with Meta-DiSc and HSROC
Meta-DiSc software and hierarchical SROC (HSROC) models combine sensitivity and specificity from different studies into summary estimates. HSROC accounts for threshold effects—differences in diagnostic cutoffs—and between-study variability. Covariate analysis can explore sources of heterogeneity, such as algorithm type or data modality.
Applications for Longevity Science
Accurate early diagnosis of infectious diseases supports global health systems and reduces long-term complications, indirectly contributing to healthier lifespans. By integrating robust ML tools into resource-limited settings, clinicians can detect and treat fevers promptly, minimizing severe disease outcomes that can impair quality of life over decades.
Further Reading
- Understanding ROC Curves: Sensitivity and Specificity Explained
- Introduction to Neural Networks for Medical Imaging
- Guide to Conducting a Diagnostic Test Accuracy Meta-Analysis