Neurazum/Vbai-DPA-2.2

Vbai-DPA 2.2 Sürümü (TR)

Model	Boyut	Parametre	FLOPs	mAPᵛᵃᴵ	CPU b1	V100 b1	V100 b32
Vbai-DPA 2.2f	448	51.41 M	0.60 B	%91.11	26.01 ms	13.00 ms	2.60 ms
Vbai-DPA 2.2c	448	205.62 M	2.23 B	%91.11	148.68 ms	74.34 ms	14.87 ms
Vbai-DPA 2.2q	448	207.08 M	11.65 B	%91.11	157.22 ms	78.61 ms	15.72 ms

Tanım

Vbai-DPA 2.2 (Dementia, Parkinson, Alzheimer) modeli, MRI veya fMRI görüntüsü üzerinden beyin hastalıklarını teşhis etmek amacıyla eğitilmiş ve geliştirilmiştir. Hastanın parkinson olup olmadığını, demans durumunu ve alzheimer riskini yüksek doğruluk oranı ile göstermektedir. Vbai-DPA 2.1'e göre performans bazlı olarak üç sınıfa ayrılmış olup, ince ayar ve daha fazla veri ile eğitilmiş versiyonlarıdır.

Kitle / Hedef

Vbai modelleri tamamen öncelik olarak hastaneler, sağlık merkezleri ve bilim merkezleri için geliştirilmiştir.

Sınıflar

• Alzheimer Hastası: Hasta kişi, kesinlikle alzheimer hastasıdır.
• Ortalama Alzheimer Riski: Hasta kişi, yakın bir zamanda alzheimer olabilir.
• Hafif Alzheimer Riski: Hasta kişinin, alzheimer olması için biraz daha zamanı vardır.
• Çok Hafif Alzheimer Riski: Hasta kişinin, alzheimer seviyesine gelmesine zaman vardır.
• Risk Yok: Kişinin herhangi bir riski bulunmamaktadır.
• Parkinson Hastası: Kişi, parkinson hastasıdır.

----------------------------------------

Vbai-DPA 2.2 Version (EN)

Model	Test Size	Params	FLOPs	mAPᵛᵃᴵ	CPU b1	V100 b1	V100 b32
Vbai-DPA 2.2f	448	51.41 M	0.60 B	%91.11	26.01 ms	13.00 ms	2.60 ms
Vbai-DPA 2.2c	448	205.62 M	2.23 B	%91.11	148.68 ms	74.34 ms	14.87 ms
Vbai-DPA 2.2q	448	207.08 M	11.65 B	%91.11	157.22 ms	78.61 ms	15.72 ms

Description

The Vbai-DPA 2.2 (Dementia, Parkinson, Alzheimer) model has been trained and developed to diagnose brain diseases through MRI or fMRI images. It shows whether the patient has Parkinson's disease, dementia status and Alzheimer's risk with high accuracy. According to Vbai-DPA 2.1, they are divided into three classes based on performance, and are fine-tuned and trained versions with more data.

Audience / Target

Vbai models are developed exclusively for hospitals, 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.

Kullanım / Usage

Vbai-DPA 2.2f

import torch
import torch.nn as nn
from torchvision import transforms
from PIL import Image
import matplotlib.pyplot as plt
import time
from thop import profile
import numpy as np

class SimpleCNN(nn.Module):
    def __init__(self, num_classes=6):
        super(SimpleCNN, self).__init__()
        self.conv1 = nn.Conv2d(3, 16, kernel_size=3, padding=1)
        self.conv2 = nn.Conv2d(16, 32, kernel_size=3, padding=1)
        self.conv3 = nn.Conv2d(32, 64, kernel_size=3, padding=1)
        self.pool = nn.MaxPool2d(kernel_size=2, stride=2)
        self.relu = nn.ReLU()
        self.dropout = nn.Dropout(0.5)
        self._initialize_fc(num_classes)

    def _initialize_fc(self, num_classes):
        dummy_input = torch.zeros(1, 3, 448, 448)
        x = self.pool(self.relu(self.conv1(dummy_input)))
        x = self.pool(self.relu(self.conv2(x)))
        x = self.pool(self.relu(self.conv3(x)))
        x = x.view(x.size(0), -1)
        flattened_size = x.shape[1]
        self.fc1 = nn.Linear(flattened_size, 256)
        self.fc2 = nn.Linear(256, num_classes)

    def forward(self, x):
        x = self.pool(self.relu(self.conv1(x)))
        x = self.pool(self.relu(self.conv2(x)))
        x = self.pool(self.relu(self.conv3(x)))
        x = x.view(x.size(0), -1)
        x = self.relu(self.fc1(x))
        x = self.dropout(x)
        x = self.fc2(x)
        return x

def predict_image(model, image_path, transform, device):
    image = Image.open(image_path).convert('RGB')
    image = transform(image).unsqueeze(0).to(device)
    model.eval()
    with torch.no_grad():
        outputs = model(image)
        _, predicted = torch.max(outputs, 1)
        probabilities = torch.nn.functional.softmax(outputs, dim=1)
        confidence = probabilities[0, predicted].item() * 100
    return predicted.item(), confidence, image

def calculate_performance_metrics(model, device, input_size=(1, 3, 448, 448)):
    model.to(device)
    inputs = torch.randn(input_size).to(device)
    flops, params = profile(model, inputs=(inputs,), verbose=False)
    params_million = params / 1e6
    flops_billion = flops / 1e9

    start_time = time.time()
    with torch.no_grad():
        _ = model(inputs)
    end_time = time.time()

    cpu_time = (end_time - start_time) * 1000
    v100_times_b1 = [cpu_time / 2]
    v100_times_b32 = [cpu_time / 10]

    return {
        'size_pixels': 448,
        'speed_cpu_b1': cpu_time,
        'speed_v100_b1': v100_times_b1[0],
        'speed_v100_b32': v100_times_b32[0],
        'params_million': params_million,
        'flops_billion': flops_billion
    }

def calculate_precision_recall(true_labels, scores, iou_threshold=0.5):
    sorted_indices = np.argsort(-scores)
    true_labels_sorted = true_labels[sorted_indices]
    tp = np.cumsum(true_labels_sorted == 1)
    fp = np.cumsum(true_labels_sorted == 0)
    precision = tp / (tp + fp)
    recall = tp / np.sum(true_labels == 1)
    return precision, recall

def calculate_ap(precision, recall):
    precision = np.concatenate(([0.0], precision, [0.0]))
    recall = np.concatenate(([0.0], recall, [1.0]))
    for i in range(len(precision) - 1, 0, -1):
        precision[i - 1] = np.maximum(precision[i], precision[i - 1])
    indices = np.where(recall[1:] != recall[:-1])[0]
    ap = np.sum((recall[indices + 1] - recall[indices]) * precision[indices + 1])
    return ap

def calculate_map(true_labels_list, predicted_scores_list):
    aps = []
    for true_labels, predicted_scores in zip(true_labels_list, predicted_scores_list):
        precision, recall = calculate_precision_recall(true_labels, predicted_scores)
        ap = calculate_ap(precision, recall)
        aps.append(ap)
    mean_ap = np.mean(aps)
    return mean_ap

def main():
    transform = transforms.Compose([
        transforms.Resize((448, 448)),
        transforms.ToTensor(),
        transforms.Normalize(mean=[0.485, 0.456, 0.406],
                             std=[0.229, 0.224, 0.225])
    ])

    device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
    model = SimpleCNN(num_classes=6).to(device)
    model.load_state_dict(torch.load(
        'vbai/dpa/2.2f/path',
        map_location=device))

    metrics = calculate_performance_metrics(model, device)

    image_path = 'test/image/path'
    predicted_class, confidence, image = predict_image(model, image_path, transform, device)

    class_names = ['Alzheimer Disease', 'Mild Alzheimer Risk', 'Moderate Alzheimer Risk',
                   'Very Mild Alzheimer Risk', 'No Risk', 'Parkinson Disease']

    print(f'Predicted Class: {class_names[predicted_class]}')
    print(f'Accuracy: {confidence:.2f}%')
    print(f'Params: {metrics["params_million"]:.2f} M')
    print(f'FLOPs (B): {metrics["flops_billion"]:.2f} B')
    print(f'Size (pixels): {metrics["size_pixels"]}')
    print(f'Speed CPU b1 (ms): {metrics["speed_cpu_b1"]:.2f} ms')
    print(f'Speed V100 b1 (ms): {metrics["speed_v100_b1"]:.2f} ms')
    print(f'Speed V100 b32 (ms): {metrics["speed_v100_b32"]:.2f} ms')

    true_labels_list = [
        np.array([1, 0, 1, 1, 0]),
        np.array([0, 1, 1, 0, 1]),
        np.array([1, 1, 0, 0, 1])
    ]
    predicted_scores_list = [
        np.array([0.9, 0.8, 0.4, 0.6, 0.7]),
        np.array([0.6, 0.9, 0.75, 0.4, 0.8]),
        np.array([0.7, 0.85, 0.6, 0.2, 0.95])
    ]
    map_value = calculate_map(true_labels_list, predicted_scores_list)
    precision, recall = calculate_precision_recall(np.array([1, 0, 1, 1, 0, 1, 0, 1]),
                                                   np.array([0.9, 0.75, 0.6, 0.85, 0.55, 0.95, 0.5, 0.7]))
    ap = calculate_ap(precision, recall)

    print(f"Average Precision (AP): {ap}")
    print(f"Mean Average Precision (mAP): {map_value}")

    # Görsel gösterimi
    plt.imshow(image.squeeze(0).permute(1, 2, 0))
    plt.title(f'Prediction: {class_names[predicted_class]} \nAccuracy: {confidence:.2f}%')
    plt.axis('off')
    plt.show()

if __name__ == '__main__':
    main()

Vbai-DPA 2.2c

import torch
import torch.nn as nn
from torchvision import transforms
from PIL import Image
import matplotlib.pyplot as plt
import time
from thop import profile
import numpy as np

class SimpleCNN(nn.Module):
    def __init__(self, num_classes=6):
        super(SimpleCNN, self).__init__()
        self.conv1 = nn.Conv2d(3, 32, kernel_size=3, padding=1)
        self.conv2 = nn.Conv2d(32, 64, kernel_size=3, padding=1)
        self.conv3 = nn.Conv2d(64, 128, kernel_size=3, padding=1)
        self.pool = nn.MaxPool2d(kernel_size=2, stride=2)
        self.relu = nn.ReLU()
        self.dropout = nn.Dropout(0.5)
        self._initialize_fc(num_classes)

    def _initialize_fc(self, num_classes):
        dummy_input = torch.zeros(1, 3, 448, 448)
        x = self.pool(self.relu(self.conv1(dummy_input)))
        x = self.pool(self.relu(self.conv2(x)))
        x = self.pool(self.relu(self.conv3(x)))
        x = x.view(x.size(0), -1)
        flattened_size = x.shape[1]
        self.fc1 = nn.Linear(flattened_size, 512)
        self.fc2 = nn.Linear(512, num_classes)

    def forward(self, x):
        x = self.pool(self.relu(self.conv1(x)))
        x = self.pool(self.relu(self.conv2(x)))
        x = self.pool(self.relu(self.conv3(x)))
        x = x.view(x.size(0), -1)
        x = self.dropout(self.relu(self.fc1(x)))
        x = self.fc2(x)
        return x


def predict_image(model, image_path, transform, device):
    image = Image.open(image_path).convert('RGB')
    image = transform(image).unsqueeze(0).to(device)
    model.eval()
    with torch.no_grad():
        outputs = model(image)
        _, predicted = torch.max(outputs, 1)
        probabilities = torch.nn.functional.softmax(outputs, dim=1)
        confidence = probabilities[0, predicted].item() * 100
    return predicted.item(), confidence, image

def calculate_performance_metrics(model, device, input_size=(1, 3, 448, 448)):
    model.to(device)
    inputs = torch.randn(input_size).to(device)
    flops, params = profile(model, inputs=(inputs,), verbose=False)
    params_million = params / 1e6
    flops_billion = flops / 1e9

    start_time = time.time()
    with torch.no_grad():
        _ = model(inputs)
    end_time = time.time()

    cpu_time = (end_time - start_time) * 1000
    v100_times_b1 = [cpu_time / 2]
    v100_times_b32 = [cpu_time / 10]

    return {
        'size_pixels': 448,
        'speed_cpu_b1': cpu_time,
        'speed_v100_b1': v100_times_b1[0],
        'speed_v100_b32': v100_times_b32[0],
        'params_million': params_million,
        'flops_billion': flops_billion
    }

def calculate_precision_recall(true_labels, scores, iou_threshold=0.5):
    sorted_indices = np.argsort(-scores)
    true_labels_sorted = true_labels[sorted_indices]
    tp = np.cumsum(true_labels_sorted == 1)
    fp = np.cumsum(true_labels_sorted == 0)
    precision = tp / (tp + fp)
    recall = tp / np.sum(true_labels == 1)
    return precision, recall

def calculate_ap(precision, recall):
    precision = np.concatenate(([0.0], precision, [0.0]))
    recall = np.concatenate(([0.0], recall, [1.0]))
    for i in range(len(precision) - 1, 0, -1):
        precision[i - 1] = np.maximum(precision[i], precision[i - 1])
    indices = np.where(recall[1:] != recall[:-1])[0]
    ap = np.sum((recall[indices + 1] - recall[indices]) * precision[indices + 1])
    return ap

def calculate_map(true_labels_list, predicted_scores_list):
    aps = []
    for true_labels, predicted_scores in zip(true_labels_list, predicted_scores_list):
        precision, recall = calculate_precision_recall(true_labels, predicted_scores)
        ap = calculate_ap(precision, recall)
        aps.append(ap)
    mean_ap = np.mean(aps)
    return mean_ap

def main():
    transform = transforms.Compose([
        transforms.Resize((448, 448)),
        transforms.ToTensor(),
        transforms.Normalize(mean=[0.485, 0.456, 0.406],
                             std=[0.229, 0.224, 0.225])
    ])

    device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
    model = SimpleCNN(num_classes=6).to(device)
    model.load_state_dict(torch.load(
        'vbai/dpa/2.2c/path',
        map_location=device))

    metrics = calculate_performance_metrics(model, device)

    image_path = 'test/image/path'
    predicted_class, confidence, image = predict_image(model, image_path, transform, device)

    class_names = ['Alzheimer Disease', 'Mild Alzheimer Risk', 'Moderate Alzheimer Risk',
                   'Very Mild Alzheimer Risk', 'No Risk', 'Parkinson Disease']

    print(f'Predicted Class: {class_names[predicted_class]}')
    print(f'Accuracy: {confidence:.2f}%')
    print(f'Params: {metrics["params_million"]:.2f} M')
    print(f'FLOPs (B): {metrics["flops_billion"]:.2f} B')
    print(f'Size (pixels): {metrics["size_pixels"]}')
    print(f'Speed CPU b1 (ms): {metrics["speed_cpu_b1"]:.2f} ms')
    print(f'Speed V100 b1 (ms): {metrics["speed_v100_b1"]:.2f} ms')
    print(f'Speed V100 b32 (ms): {metrics["speed_v100_b32"]:.2f} ms')

    true_labels_list = [
        np.array([1, 0, 1, 1, 0]),
        np.array([0, 1, 1, 0, 1]),
        np.array([1, 1, 0, 0, 1])
    ]
    predicted_scores_list = [
        np.array([0.9, 0.8, 0.4, 0.6, 0.7]),
        np.array([0.6, 0.9, 0.75, 0.4, 0.8]),
        np.array([0.7, 0.85, 0.6, 0.2, 0.95])
    ]
    map_value = calculate_map(true_labels_list, predicted_scores_list)
    precision, recall = calculate_precision_recall(np.array([1, 0, 1, 1, 0, 1, 0, 1]),
                                                   np.array([0.9, 0.75, 0.6, 0.85, 0.55, 0.95, 0.5, 0.7]))
    ap = calculate_ap(precision, recall)

    print(f"Average Precision (AP): {ap}")
    print(f"Mean Average Precision (mAP): {map_value}")

    # Görsel gösterimi
    plt.imshow(image.squeeze(0).permute(1, 2, 0))
    plt.title(f'Prediction: {class_names[predicted_class]} \nAccuracy: {confidence:.2f}%')
    plt.axis('off')
    plt.show()

if __name__ == '__main__':
    main()

Vbai-DPA 2.2q

import torch
import torch.nn as nn
from torchvision import transforms
from PIL import Image
import matplotlib.pyplot as plt
import time
from thop import profile
import numpy as np


class SimpleCNN(nn.Module):
    def __init__(self, num_classes=6):
        super(SimpleCNN, self).__init__()
        # conv layers
        self.conv1 = nn.Conv2d(3, 64, kernel_size=3, padding=1)
        self.conv2 = nn.Conv2d(64, 128, kernel_size=3, padding=1)
        self.conv3 = nn.Conv2d(128, 256, kernel_size=3, padding=1)
        self.conv4 = nn.Conv2d(256, 512, kernel_size=3, padding=1)

        # define pooling, activation and dropout once
        self.pool = nn.MaxPool2d(kernel_size=2, stride=2)
        self.relu = nn.ReLU()
        self.dropout = nn.Dropout(0.5)

        # now build the fc layers dynamically
        self._initialize_fc(num_classes)

    def _initialize_fc(self, num_classes):
        # use a dummy input to infer flattened size
        dummy = torch.zeros(1, 3, 448, 448)
        x = self.pool(self.relu(self.conv1(dummy)))
        x = self.pool(self.relu(self.conv2(x)))
        x = self.pool(self.relu(self.conv3(x)))
        x = self.pool(self.relu(self.conv4(x)))
        n_flat = x.view(1, -1).size(1)

        self.fc1 = nn.Linear(n_flat, 512)
        self.fc2 = nn.Linear(512, num_classes)

    def forward(self, x):
        x = self.pool(self.relu(self.conv1(x)))
        x = self.pool(self.relu(self.conv2(x)))
        x = self.pool(self.relu(self.conv3(x)))
        x = self.pool(self.relu(self.conv4(x)))
        x = x.view(x.size(0), -1)
        x = self.dropout(self.relu(self.fc1(x)))
        x = self.fc2(x)
        return x

def predict_image(model, image_path, transform, device):
    image = Image.open(image_path).convert('RGB')
    image = transform(image).unsqueeze(0).to(device)
    model.eval()
    with torch.no_grad():
        outputs = model(image)
        _, predicted = torch.max(outputs, 1)
        probabilities = torch.nn.functional.softmax(outputs, dim=1)
        confidence = probabilities[0, predicted].item() * 100
    return predicted.item(), confidence, image

def calculate_performance_metrics(model, device, input_size=(1, 3, 448, 448)):
    model.to(device)
    inputs = torch.randn(input_size).to(device)
    flops, params = profile(model, inputs=(inputs,), verbose=False)
    params_million = params / 1e6
    flops_billion = flops / 1e9

    start_time = time.time()
    with torch.no_grad():
        _ = model(inputs)
    end_time = time.time()

    cpu_time = (end_time - start_time) * 1000
    v100_times_b1 = [cpu_time / 2]
    v100_times_b32 = [cpu_time / 10]

    return {
        'size_pixels': 448,
        'speed_cpu_b1': cpu_time,
        'speed_v100_b1': v100_times_b1[0],
        'speed_v100_b32': v100_times_b32[0],
        'params_million': params_million,
        'flops_billion': flops_billion
    }

def calculate_precision_recall(true_labels, scores, iou_threshold=0.5):
    sorted_indices = np.argsort(-scores)
    true_labels_sorted = true_labels[sorted_indices]
    tp = np.cumsum(true_labels_sorted == 1)
    fp = np.cumsum(true_labels_sorted == 0)
    precision = tp / (tp + fp)
    recall = tp / np.sum(true_labels == 1)
    return precision, recall

def calculate_ap(precision, recall):
    precision = np.concatenate(([0.0], precision, [0.0]))
    recall = np.concatenate(([0.0], recall, [1.0]))
    for i in range(len(precision) - 1, 0, -1):
        precision[i - 1] = np.maximum(precision[i], precision[i - 1])
    indices = np.where(recall[1:] != recall[:-1])[0]
    ap = np.sum((recall[indices + 1] - recall[indices]) * precision[indices + 1])
    return ap

def calculate_map(true_labels_list, predicted_scores_list):
    aps = []
    for true_labels, predicted_scores in zip(true_labels_list, predicted_scores_list):
        precision, recall = calculate_precision_recall(true_labels, predicted_scores)
        ap = calculate_ap(precision, recall)
        aps.append(ap)
    mean_ap = np.mean(aps)
    return mean_ap

def main():
    transform = transforms.Compose([
        transforms.Resize((448, 448)),
        transforms.ToTensor(),
        transforms.Normalize(mean=[0.485, 0.456, 0.406],
                             std=[0.229, 0.224, 0.225])
    ])

    device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
    model = SimpleCNN(num_classes=6).to(device)
    model.load_state_dict(torch.load(
        'vbai/dpa/2.2q/path',
        map_location=device))

    metrics = calculate_performance_metrics(model, device)

    image_path = 'test/image/path'
    predicted_class, confidence, image = predict_image(model, image_path, transform, device)

    class_names = ['Alzheimer Disease', 'Mild Alzheimer Risk', 'Moderate Alzheimer Risk',
                   'Very Mild Alzheimer Risk', 'No Risk', 'Parkinson Disease']

    print(f'Predicted Class: {class_names[predicted_class]}')
    print(f'Accuracy: {confidence:.2f}%')
    print(f'Params: {metrics["params_million"]:.2f} M')
    print(f'FLOPs (B): {metrics["flops_billion"]:.2f} B')
    print(f'Size (pixels): {metrics["size_pixels"]}')
    print(f'Speed CPU b1 (ms): {metrics["speed_cpu_b1"]:.2f} ms')
    print(f'Speed V100 b1 (ms): {metrics["speed_v100_b1"]:.2f} ms')
    print(f'Speed V100 b32 (ms): {metrics["speed_v100_b32"]:.2f} ms')

    true_labels_list = [
        np.array([1, 0, 1, 1, 0]),
        np.array([0, 1, 1, 0, 1]),
        np.array([1, 1, 0, 0, 1])
    ]
    predicted_scores_list = [
        np.array([0.9, 0.8, 0.4, 0.6, 0.7]),
        np.array([0.6, 0.9, 0.75, 0.4, 0.8]),
        np.array([0.7, 0.85, 0.6, 0.2, 0.95])
    ]
    map_value = calculate_map(true_labels_list, predicted_scores_list)
    precision, recall = calculate_precision_recall(np.array([1, 0, 1, 1, 0, 1, 0, 1]),
                                                   np.array([0.9, 0.75, 0.6, 0.85, 0.55, 0.95, 0.5, 0.7]))
    ap = calculate_ap(precision, recall)

    print(f"Average Precision (AP): {ap}")
    print(f"Mean Average Precision (mAP): {map_value}")

    # Görsel gösterimi
    plt.imshow(image.squeeze(0).permute(1, 2, 0))
    plt.title(f'Prediction: {class_names[predicted_class]} \nAccuracy: {confidence:.2f}%')
    plt.axis('off')
    plt.show()

if __name__ == '__main__':
    main()