Create README.md

Brain Tumor Detection Model Training
/

Beyin Tümörü Tespiti Model Eğitimi
Proje Açıklaması / Project Description
Bu proje, beyin tümörü MRI görüntülerini sınıflandırmak için çeşitli derin öğrenme modellerini kullanarak transfer öğrenme ve özel CNN mimarisi ile eğitim yapmaktadır. Model eğitimleri sonunda TensorFlow Lite formatında çıktı alınmış ve eğitim süreçlerine ait performans grafikleri ile konfüzyon matrisleri oluşturulmuştur.
This project trains various deep learning models using transfer learning and custom CNN architecture to classify brain tumor MRI images. At the end of training, models are converted to TensorFlow Lite format, and training performance graphs along with confusion matrices are generated.



Kullanılan Modeller / Models Used
MobileNetV2 (Transfer Learning)

ResNet50 (Transfer Learning)




Eğitim Parametreleri / Training Parameters

| Parametre / Parameter | Değer / Value |
| ---------------------------- | ------------- |
| Görüntü Boyutu / Image Size | 224x224 |
| Batch Size | 32 |
| Epoch Sayısı / Epochs | 15 |
| Öğrenme Hızı / Learning Rate | 0.0001 (1e-4) |



Eğitim Süreci / Training Process
Veriler dataset/Training ve dataset/Testing klasörlerinden okunmuştur.

Data augmentation uygulanarak (döndürme, zoom, yatay ve dikey çevirme) eğitim seti çeşitlendirilmiştir.

Transfer öğrenme modellerinde son %20 katman serbest bırakılarak eğitim yapılmıştır.

EarlyStopping ve ReduceLROnPlateau callback’leri kullanılarak aşırı öğrenme önlenmiş ve öğrenme hızı dinamik olarak ayarlanmıştır.

Her model için eğitim kayıpları ve doğrulukları grafik olarak kaydedilmiştir.

Test verisinde F1 skorları ve sınıflandırma raporları hesaplanmıştır.

Eğitilen modeller TensorFlow Lite formatına dönüştürülüp tflite_models klasörüne kaydedilmiştir.



Kullanılan Kütüphaneler / Libraries Used
TensorFlow

NumPy

scikit-learn

Matplotlib
















README.md

@@ -0,0 +1,75 @@
+---
+license: mit
+language:
+- tr
+- en
+metrics:
+- accuracy
+- f1
+- confusion_matrix
+base_model:
+- microsoft/resnet-50
+pipeline_tag: image-classification
+tags:
+- biology
+---
+Model Summary
+This model card describes two deep learning models trained to classify brain tumor MRI images into different tumor types. The models are based on ResNet50 and MobileNetV2 architectures and were trained using the Brain Tumor MRI Dataset available on Kaggle. They aim to assist medical professionals in detecting brain tumors using transfer learning approaches.
+Model Details
+ResNet50
+
+- Developed by: [Your Name or Organization]
+- Model type: Convolutional Neural Network (CNN)
+- Language(s): N/A (Image classification task)
+- License: [Specify license, e.g., MIT, Apache 2.0]
+- Finetuned from model: ResNet50 (ImageNet pretrained)
+
+MobileNetV2
+
+- Developed by: [Your Name or Organization]
+- Model type: Convolutional Neural Network (CNN)
+- Language(s): N/A (Image classification task)
+- License: [Specify license, e.g., MIT, Apache 2.0]
+- Finetuned from model: MobileNetV2 (ImageNet pretrained)
+
+Uses
+Direct Use
+These models can be used to classify MRI scans for brain tumor detection in clinical decision-support systems.
+Out-of-Scope Use
+Not intended for standalone diagnostic purposes without medical supervision. Misuse includes deployment without validation or interpretability assessments.
+Bias, Risks, and Limitations
+The model performance may vary depending on the image quality, scanner differences, and patient demographics. Models may inherit biases from the training data.
+Training Details
+Training Data
+The models were trained on the Brain Tumor MRI Dataset from Kaggle. The dataset contains images categorized into three classes: glioma, meningioma, and pituitary tumors.
+Training Hyperparameters
+
+- Learning Rate: 0.0001
+- Epochs: 25
+- Batch Size: 32
+- Optimizer: Adam
+- Loss Function: Categorical Cross-Entropy
+
+Evaluation
+Testing Data
+Test data is a stratified split of the original dataset with unseen examples from each tumor class.
+Metrics
+Accuracy, Precision, Recall, F1-score
+Results
+
+- ResNet50 Accuracy: ~98%
+- MobileNetV2 Accuracy: ~96%
+
+Environmental Impact
+
+- Hardware Type: NVIDIA GPU (e.g., RTX 3060 or Colab T4)
+- Hours Used: ~1-2 hours per model
+- Cloud Provider: Google Colab
+- Compute Region: [Not specified]
+- Carbon Emitted: Estimated using ML Impact calculator (minimal due to short training duration)
+
+Technical Specifications
+Model Architecture and Objective
+Transfer learning using pretrained CNNs (ResNet50 and MobileNetV2) adapted for multi-class classification of brain tumor MRI images.
+Software
+Python 3.10, TensorFlow/Keras, NumPy, Matplotlib, OpenCV