Vbai 2.5

Vbai-2.5

General Tests

Model	Test Size	Params	Accuracy	mAPvaI	F1 Score	Recall	Precision
Vbai-2.5f	224	129M	71.71%	72.41%	71.21%	69.58%	76.64%
Vbai-2.5q	224	220M	82.44%	84.40%	80.95%	83.99%	80.08%
Vbai-2.5T+	224	207M	65.55%	67.33%	59.82%	67.62%	63.56%

Vbai-2.5T+ is a version fine-tuned for tumor classification. It has low accuracy for atrophy and dementia.

Dementia-Based Tests

Model	Accuracy	mAPvaI	F1 Score	Recall
Vbai-2.5f	71.71%	72.41%	71.21%	69.58%
Vbai-2.5q	82.44%	84.40%	80.95%	83.99%
Vbai-2.5T+	35.86%	35.54%	24.38%	40.00%

Tumor-Based Tests

Model	Accuracy	mAPvaI	F1 Score	Recall
Vbai-2.5f	95.48%	98.78%	95.49%	95.48%
Vbai-2.5q	97.14%	99.65%	97.15%	97.14%
Vbai-2.5T+	95.24%	99.12%	95.26%	95.24%

Description

The Vbai-2.5 model has been trained and developed to diagnose brain diseases using MRI or fMRI images. It indicates whether the patient has Parkinson's disease, dementia, and Alzheimer's risk, and shows tumor stages with a high accuracy rate. Unlike other Vbai models, its diagnostic capability is now consolidated into a single model.

This multi-task model also serves as the foundation of an image processing library. It can run and diagnose both classes separately or simultaneously.

Audience / Target

Vbai models are developed exclusively for hospitals, universities, communities, health centres and science centres.

Classes

• Alzheimer's disease: The sick person definitely has Alzheimer's disease.
• Average Risk of Alzheimer's Disease: The sick person may develop Alzheimer's disease in the near future.
• Mild Alzheimer's Risk: The sick person has a little more time to develop Alzheimer's disease.
• Very Mild Alzheimer's Risk: The sick person has time to reach the level of Alzheimer's disease.
• No Risk: The person does not have any risk.
• Parkinson's Disease: The person has Parkinson's disease.
• Glioma Tumor: The tumor in the person has an aggressive behavior.
• Meningioma Tumor: The tumor is progressing.
• Pituitary Tumor: The tumor in the person is progressing slowly.
• No Tumor: The person does not have a tumor.

Usage

import torch
import torch.nn as nn
import torchvision.transforms as transforms
from PIL import Image
import matplotlib.pyplot as plt
import numpy as np
import seaborn as sns
import cv2
import os
import torch.nn.functional as F
from sklearn.metrics import classification_report, confusion_matrix

DEVICE = torch.device("cuda" if torch.cuda.is_available() else "cpu")
MODEL_PATH = 'Vbai-2.5/model/path'
MODEL_TYPE = 'f'

DEMENTIA_DIR = 'dementia/data/dir'
TUMOR_DIR = 'tumor/data/dir'

TEST_IMAGE_PATH = 'single/test/image/file'

DEM_CLASSES = ['AD Alzheimer Diseases', 'AD Mild Demented', 'AD Moderate Demented',
               'AD Very Mild Demented', 'CN Non Demented', 'PD Parkinson Diseases']
TUM_CLASSES = ['Glioma Tumor', 'Meningioma Tumor', 'No Tumor', 'Pituitary Tumor']


# --- MODEL ARCHITECTURE ---
class AttentionModule(nn.Module):
    def __init__(self, in_channels):
        super(AttentionModule, self).__init__()
        self.conv1 = nn.Conv2d(in_channels, max(1, in_channels // 8), 1)
        self.conv2 = nn.Conv2d(max(1, in_channels // 8), 1, 1)
        self.sigmoid = nn.Sigmoid()

    def forward(self, x):
        att = self.sigmoid(self.conv2(F.relu(self.conv1(x))))
        return x * att, att


class MultiTaskBrainModel(nn.Module):
    def __init__(self, num_dem_classes, num_tum_classes, model_type='f'):
        super(MultiTaskBrainModel, self).__init__()
        self.model_type = model_type

        if model_type == 'f':
            self.conv1 = nn.Conv2d(3, 32, 3, padding=1)
            self.conv2 = nn.Conv2d(32, 64, 3, padding=1)
            self.conv3 = nn.Conv2d(64, 128, 3, padding=1)
            final_ch = 128
        elif model_type == 'q':
            self.conv1 = nn.Conv2d(3, 64, 3, padding=1)
            self.bn1 = nn.BatchNorm2d(64)
            self.conv2 = nn.Conv2d(64, 128, 3, padding=1)
            self.bn2 = nn.BatchNorm2d(128)
            self.conv3 = nn.Conv2d(128, 256, 3, padding=1)
            self.bn3 = nn.BatchNorm2d(256)
            self.conv4 = nn.Conv2d(256, 512, 3, padding=1)
            self.bn4 = nn.BatchNorm2d(512)
            final_ch = 512

        self.pool = nn.MaxPool2d(2, 2)
        self.relu = nn.ReLU()
        self.dropout = nn.Dropout(0.5 if model_type == 'q' else 0.3)

        self.edge_conv1 = nn.Conv2d(1, 16, 3, padding=1)
        self.edge_conv2 = nn.Conv2d(16, 32, 3, padding=1)
        self.edge_pool = nn.AdaptiveAvgPool2d(28)

        self.dem_att = AttentionModule(final_ch)
        self.tum_att = AttentionModule(final_ch)

        if model_type == 'q':
            self.feat_dim = final_ch * 14 * 14
            self.edge_dim = 32 * 14 * 14
            self.edge_pool = nn.AdaptiveAvgPool2d(14)
        else:
            self.feat_dim = final_ch * 28 * 28
            self.edge_dim = 32 * 28 * 28

        self.dem_fc = nn.Sequential(
            nn.Linear(self.feat_dim + self.edge_dim, 512 if model_type == 'f' else 1024),
            nn.ReLU(),
            nn.Dropout(0.4),
            nn.Linear(512 if model_type == 'f' else 1024, num_dem_classes)
        )

        self.tum_fc = nn.Sequential(
            nn.Linear(self.feat_dim + self.edge_dim, 512 if model_type == 'f' else 1024),
            nn.ReLU(),
            nn.Dropout(0.4),
            nn.Linear(512 if model_type == 'f' else 1024, num_tum_classes)
        )

    def forward(self, x, edge_x):
        if self.model_type == 'f':
            x = self.pool(self.relu(self.conv1(x)))
            x = self.pool(self.relu(self.conv2(x)))
            x = self.relu(self.conv3(x))
            x = self.pool(x)
        else:
            x = self.pool(self.relu(self.bn1(self.conv1(x))))
            x = self.pool(self.relu(self.bn2(self.conv2(x))))
            x = self.pool(self.relu(self.bn3(self.conv3(x))))
            x = self.relu(self.bn4(self.conv4(x)))
            x = self.pool(x)

        e = self.pool(self.relu(self.edge_conv1(edge_x)))
        e = self.relu(self.edge_conv2(e))
        e = self.edge_pool(e)
        e_flat = e.view(e.size(0), -1)

        d_x, d_att = self.dem_att(x)
        d_flat = d_x.view(d_x.size(0), -1)
        dem_out = self.dem_fc(torch.cat([d_flat, e_flat], dim=1))

        t_x, t_att = self.tum_att(x)
        t_flat = t_x.view(t_x.size(0), -1)
        tum_out = self.tum_fc(torch.cat([t_flat, e_flat], dim=1))

        return dem_out, tum_out, d_att, t_att


transform = transforms.Compose([
    transforms.Resize((224, 224)),
    transforms.ToTensor(),
    transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
])

edge_transform = transforms.Compose([
    transforms.Resize((224, 224)),
    transforms.Grayscale(num_output_channels=1),
    transforms.ToTensor()
])


def analyze_single_image(model, image_path):
    print(f"\nIs being analyzed: {image_path}")
    if not os.path.exists(image_path):
        print("File not found.")
        return

    try:
        raw_image = Image.open(image_path).convert('RGB')
        img_tensor = transform(raw_image).unsqueeze(0).to(DEVICE)
        edge_tensor = edge_transform(raw_image).unsqueeze(0).to(DEVICE)

        model.eval()
        with torch.no_grad():
            d_out, t_out, d_att, t_att = model(img_tensor, edge_tensor)

            d_prob = F.softmax(d_out, dim=1)
            t_prob = F.softmax(t_out, dim=1)

            d_pred = torch.argmax(d_prob).item()
            t_pred = torch.argmax(t_prob).item()

        print("-" * 50)
        print(f"DEMENTIA: {DEM_CLASSES[d_pred]} (%{d_prob[0][d_pred] * 100:.1f})")
        print(f"TUMOR : {TUM_CLASSES[t_pred]} (%{t_prob[0][t_pred] * 100:.1f})")
        print("-" * 50)

        d_map = d_att.detach().cpu().numpy()[0, 0]
        t_map = t_att.detach().cpu().numpy()[0, 0]

        d_map = cv2.resize(d_map, (224, 224))
        t_map = cv2.resize(t_map, (224, 224))

        plt.figure(figsize=(12, 4))

        plt.subplot(1, 3, 1)
        plt.imshow(raw_image)
        plt.title("Original Image")
        plt.axis('off')

        plt.subplot(1, 3, 2)
        plt.imshow(raw_image)
        plt.imshow(d_map, cmap='Blues', alpha=0.5)
        plt.title(f"Dementia Attention\n{DEM_CLASSES[d_pred]}")
        plt.axis('off')

        plt.subplot(1, 3, 3)
        plt.imshow(raw_image)
        plt.imshow(t_map, cmap='Reds', alpha=0.5)
        plt.title(f"Tumor Attention\n{TUM_CLASSES[t_pred]}")
        plt.axis('off')

        plt.tight_layout()
        plt.show()

    except Exception as e:
        print(f"Error: {e}")
        import traceback
        traceback.print_exc()


def evaluate_dataset_performance(model, dementia_dir, tumor_dir):
    print("\n" + "=" * 60)
    print("Starting the performance test")
    print("=" * 60)

    model.eval()

    print(f"\nDementia data set: {dementia_dir}")
    d_true, d_pred = [], []

    if os.path.exists(dementia_dir):
        for idx, cls_name in enumerate(DEM_CLASSES):
            cls_path = os.path.join(dementia_dir, cls_name)
            if not os.path.exists(cls_path): continue

            files = [f for f in os.listdir(cls_path) if f.lower().endswith(('.jpg', '.png'))][:50]

            for fname in files:
                try:
                    img = Image.open(os.path.join(cls_path, fname)).convert('RGB')
                    img_t = transform(img).unsqueeze(0).to(DEVICE)
                    edge_t = edge_transform(img).unsqueeze(0).to(DEVICE)
                    with torch.no_grad():
                        out, _, _, _ = model(img_t, edge_t)
                        d_true.append(idx)
                        d_pred.append(torch.argmax(out, 1).item())
                except:
                    pass

        if d_true:
            print(classification_report(d_true, d_pred, target_names=DEM_CLASSES, digits=3))

    print(f"\nTumor dataset: {tumor_dir}")
    t_true, t_pred = [], []

    if os.path.exists(tumor_dir):
        for idx, cls_name in enumerate(TUM_CLASSES):
            cls_path = os.path.join(tumor_dir, cls_name)
            if not os.path.exists(cls_path): continue

            files = [f for f in os.listdir(cls_path) if f.lower().endswith(('.jpg', '.png'))][:50]

            for fname in files:
                try:
                    img = Image.open(os.path.join(cls_path, fname)).convert('RGB')
                    img_t = transform(img).unsqueeze(0).to(DEVICE)
                    edge_t = edge_transform(img).unsqueeze(0).to(DEVICE)
                    with torch.no_grad():
                        _, out, _, _ = model(img_t, edge_t)
                        t_true.append(idx)
                        t_pred.append(torch.argmax(out, 1).item())
                except:
                    pass

        if t_true:
            print(classification_report(t_true, t_pred, target_names=TUM_CLASSES, digits=3))


def main():
    print("System loading...")
    model = MultiTaskBrainModel(len(DEM_CLASSES), len(TUM_CLASSES), model_type=MODEL_TYPE).to(DEVICE)

    try:
        model.load_state_dict(torch.load(MODEL_PATH, map_location=DEVICE))...