Vbai 2.6AD

--- license: cc-by-nc-sa-4.0 language: - en - tr pipeline_tag: image-classification library_name: pytorch metrics: - f1 - accuracy - precision - recall tags: - research - medical - health - brain - mri - diagnosis - neuroscience - doctor - healthcare - hospital - academic - professional - alzheimer - athropy - dementia - early-diagnosis - biomarkers - blood - MCI - CN - AD - multimodal - tabular - brain-imaging --- # Vbai-2.6AD ## Description Vbai-2.6 is a 3D brain MRI segmentation model developed as the latest generation member of the Vbai model family. Unlike previous versions, Vbai-2.6AD is now a multimodal model designed for professional research purposes, capable of working with NIfTI files and blood values. Vbai-3D versions have been integrated with the standard Vbai versions. The model generates voxel-level segmentation masks instead of image-level labels and provides the spatial localization of pathological regions in addition to quantitative tissue volume measurements. It aims to accurately establish a diagnosis by establishing a context with blood values and patient information. Vbai-2.6AD also serves as the core engine of the HealFuture image processing library and can run each diagnostic task independently or in combination, depending on the clinical use case. This model has been trained exclusively for the early diagnosis of atrophy and Alzheimer’s disease. ### Audience / Target Vbai models are developed exclusively for hospitals, universities, communities, health centres and science centres. ## Architecture This model is a multimodal deep learning architecture that combines 3D MRI data and clinical biomarkers for Alzheimer's classification and progression prediction. <table> <tr> <th>Input Modality</th> <th>Encoder</th> <th>Fusion Layer</th> <th>Prediction Heads (Output)</th> </tr> <tr> <td><b>3D MRI Volume</b><br>(96x96x96, 1ch)</td> <td>3D ResNet + CBAM<br>+ SE Blocks + ASPP</td> <td rowspan="2"><b>Bidirectional<br>Cross-Attention</b><br>(MRI ↔ Tabular)<br>+ Gated Residuals</td> <td>→ Diagnosis: <b>CN / MCI / AD Classification</b></td> </tr> <tr> <td><b>Clinical Data</b><br>(13 Features + Masks)</td> <td>MLP (LayerNorm + GELU)<br>Linear Projection</td> <td>→ Progression: <b>MCI to AD Conversion & Timeline Estimation</b></td> </tr> <tr> <td colspan="2" align="center"><i>Auxiliary Supervision (Training Mode)</i><...