A hybrid ViT-L/32–MaxViT-L architecture with adaptive gated fusion for multiclass gastrointestinal disease detection and multi-method post-hoc explainability - Scorecard - MDSpire
Advertisement
A hybrid ViT-L/32–MaxViT-L architecture with adaptive gated fusion for multiclass gastrointestinal disease detection and multi-method post-hoc explainability
Clinical Scorecard: An Adaptive Gated Fusion Approach Utilizing a Hybrid ViT-L/32–MaxViT-L Model for Multiclass Detection of Gastrointestinal Disorders and Enhanced Post-Hoc Explainability
At a Glance
Category
Detail
Condition
Gastrointestinal Disorders
Key Mechanisms
Hybrid transformer architecture integrating global token-level reasoning and hierarchical spatial modeling with adaptive gated fusion.
Target Population
Patients undergoing endoscopic imaging for gastrointestinal disease detection.
Care Setting
Clinical practice involving gastrointestinal screening and diagnosis.
Key Highlights
Proposed framework outperformed state-of-the-art transformer and hybrid architectures in multiclass gastrointestinal disease detection.
Demonstrated high accuracy, sensitivity, specificity, and area under the ROC curve across eight disease categories.
Adaptive gated fusion enhances diagnostic performance and interpretability.
Guideline-Based Recommendations
Diagnosis
Utilize endoscopic imaging as the primary diagnostic modality for gastrointestinal diseases.
Management
Implement computer-aided diagnostic systems to improve consistency and sensitivity in gastrointestinal screening.
Monitoring & Follow-up
Employ explainable AI techniques to enhance interpretability of diagnostic decisions.
Risks
Operator-dependent interpretation may lead to inter-observer variability and misclassification of lesions.
Patient & Prescribing Data
Individuals with suspected gastrointestinal disorders requiring endoscopic evaluation.
Timely and accurate detection of lesions is critical for effective clinical management.
Clinical Best Practices
Incorporate hybrid transformer models for enhanced diagnostic accuracy in gastrointestinal imaging.
Utilize adaptive mechanisms to balance local and global feature representations.
