Merge pull request #753 from Arihant-Bhandari/fracture

[Project Addition] Fracture Detection using DL
abhisheks008 · Jun 7, 2024 · ee7b263 · ee7b263
2 parents f2ba2d7 + 8ad07f7
commit ee7b263
Show file tree

Hide file tree

Showing 20 changed files with 155 additions and 0 deletions.
diff --git a/Fracture Detection using DL/Dataset/README.md b/Fracture Detection using DL/Dataset/README.md
@@ -0,0 +1,12 @@
+The link for the dataset used in this project:  https://www.kaggle.com/datasets/bmadushanirodrigo/fracture-multi-region-x-ray-data
+
+The dataset consists of 3 subdirectories under the bone_fracture_binary_classification directory, train, test and val, all three with 2 subdirectories: fracture and not-fractured; train with approximately 9200 images, val with approximately 850 and test with approximately 600 images respectively.
+
+
+**Appropriate image count**
+
+Train images images: 9165
+
+Validation images: 764
+
+Test images: 443
diff --git a/Fracture Detection using DL/Images/CNN Accuracy.png b/Fracture Detection using DL/Images/CNN Accuracy.png
diff --git a/Fracture Detection using DL/Images/CNN Loss.png b/Fracture Detection using DL/Images/CNN Loss.png
diff --git a/Fracture Detection using DL/Images/DenseNet121 Accuracy.png b/Fracture Detection using DL/Images/DenseNet121 Accuracy.png
diff --git a/Fracture Detection using DL/Images/DenseNet121 Loss.png b/Fracture Detection using DL/Images/DenseNet121 Loss.png
diff --git a/Fracture Detection using DL/Images/EDA.png b/Fracture Detection using DL/Images/EDA.png
diff --git a/Fracture Detection using DL/Images/Fracture.png b/Fracture Detection using DL/Images/Fracture.png
diff --git a/Fracture Detection using DL/Images/InceptionV3 Accuracy.png b/Fracture Detection using DL/Images/InceptionV3 Accuracy.png
diff --git a/Fracture Detection using DL/Images/InceptionV3 Loss.png b/Fracture Detection using DL/Images/InceptionV3 Loss.png
diff --git a/Fracture Detection using DL/Images/MobileNet Accuracy.png b/Fracture Detection using DL/Images/MobileNet Accuracy.png
diff --git a/Fracture Detection using DL/Images/MobileNet Loss.png b/Fracture Detection using DL/Images/MobileNet Loss.png
diff --git a/Fracture Detection using DL/Images/Not fractured.png b/Fracture Detection using DL/Images/Not fractured.png
diff --git a/Fracture Detection using DL/Images/RESNET50 Accuracy.png b/Fracture Detection using DL/Images/RESNET50 Accuracy.png
diff --git a/Fracture Detection using DL/Images/RESNET50 Loss.png b/Fracture Detection using DL/Images/RESNET50 Loss.png
diff --git a/Fracture Detection using DL/Images/VGG16 Accuracy.png b/Fracture Detection using DL/Images/VGG16 Accuracy.png
diff --git a/Fracture Detection using DL/Images/VGG16 Loss.png b/Fracture Detection using DL/Images/VGG16 Loss.png
diff --git a/Fracture Detection using DL/Images/Xception Accuracy.png b/Fracture Detection using DL/Images/Xception Accuracy.png
diff --git a/Fracture Detection using DL/Images/Xception Loss.png b/Fracture Detection using DL/Images/Xception Loss.png
diff --git a/Fracture Detection using DL/Model/fracture-classify.ipynb b/Fracture Detection using DL/Model/fracture-classify.ipynb
diff --git a/Fracture Detection using DL/README.md b/Fracture Detection using DL/README.md
@@ -0,0 +1,142 @@
+## **Fracture Detection using DL**
+
+### 🎯 **Goal**
+
+The objective of this project is to classify images of x-ray scans of bones, into 2 classes: fracture and not-fractured.
+
+### 🧵 **Dataset**
+
+The dataset consists of 3 subdirectories under the bone_fracture_binary_classification directory, train, test and val, all three with 2 subdirectories: fracture and not-fractured; train with approximately 9200 images, val with approximately 850 and test with approximately 600 images respectively.
+
+**Appropriate image count**
+
+Train images images: 9165
+
+Validation images: 764
+
+Test images: 443
+
+### 🧾 **Description**
+
+The project deals with binary classification, classifying images into 2 categories: fracture and not-fractured bone x-ray scans.
+
+### 🧮 **What I had done!**
+
+To achieve our goals, the following steps were implemented:
+
+- Images were loaded using keras.utils and normalized to the range 0 to 1.
+
+- Images were scanned for appropriateness.
+
+- Images were resized to a fixed size of 224x224 pixels.
+
+- Custom and pre-trained models were used for this task.
+
+### 🚀 **Models Implemented**
+
+models used:
+
+- ResNet-50
+- Xception
+- VGG16
+- CNN
+- InceptionV3
+- DenseNet-121
+- MobileNet
+
+### 📚 **Libraries Needed**
+
+- Keras
+
+- Tensorflow
+
+- Numpy
+
+- Matplotlib
+
+### 📊 **Exploratory Data Analysis Results**
+
+
+- #### **EDA**
+
+<p align="center">
+  <img src="https://github.com/Arihant-Bhandari/DL-Simplified/blob/fracture/Fracture%20Detection%20using%20DL/Images/Fracture.png" height="400px" width="400px" />
+  <img src="https://github.com/Arihant-Bhandari/DL-Simplified/blob/fracture/Fracture%20Detection%20using%20DL/Images/Not%20fractured.png" height="400px" width="400px" />
+</p>
+
+<img src="https://github.com/Arihant-Bhandari/DL-Simplified/blob/fracture/Fracture%20Detection%20using%20DL/Images/EDA.png"/>
+
+- #### **DenseNet-121**
+
+<p align="center">
+  <img src="https://github.com/Arihant-Bhandari/DL-Simplified/blob/fracture/Fracture%20Detection%20using%20DL/Images/DenseNet121%20Accuracy.png" height="400px" width="400px" />
+  <img src="https://github.com/Arihant-Bhandari/DL-Simplified/blob/fracture/Fracture%20Detection%20using%20DL/Images/DenseNet121%20Loss.png" height="400px" width="400px" />
+</p>
+
+- #### **CNN**
+
+<p align="center">
+  <img src="https://github.com/Arihant-Bhandari/DL-Simplified/blob/fracture/Fracture%20Detection%20using%20DL/Images/CNN%20Accuracy.png" height="400px" width="400px" />
+  <img src="https://github.com/Arihant-Bhandari/DL-Simplified/blob/fracture/Fracture%20Detection%20using%20DL/Images/CNN%20Loss.png" height="400px" width="400px" />
+</p>
+
+- #### **InceptionV3**
+
+<p align="center">
+  <img src="https://github.com/Arihant-Bhandari/DL-Simplified/blob/fracture/Fracture%20Detection%20using%20DL/Images/InceptionV3%20Accuracy.png" height="400px" width="400px" />
+  <img src="https://github.com/Arihant-Bhandari/DL-Simplified/blob/fracture/Fracture%20Detection%20using%20DL/Images/InceptionV3%20Loss.png" height="400px" width="400px" />
+</p>
+
+- #### **VGG16**
+
+<p align="center">
+  <img src="https://github.com/Arihant-Bhandari/DL-Simplified/blob/fracture/Fracture%20Detection%20using%20DL/Images/VGG16%20Accuracy.png" height="400px" width="400px" />
+  <img src="https://github.com/Arihant-Bhandari/DL-Simplified/blob/fracture/Fracture%20Detection%20using%20DL/Images/VGG16%20Loss.png" height="400px" width="400px" />
+</p>
+
+- #### **MobileNet**
+
+<p align="center">
+  <img src="https://github.com/Arihant-Bhandari/DL-Simplified/blob/fracture/Fracture%20Detection%20using%20DL/Images/MobileNet%20Accuracy.png" height="400px" width="400px" />
+  <img src="https://github.com/Arihant-Bhandari/DL-Simplified/blob/fracture/Fracture%20Detection%20using%20DL/Images/MobileNet%20Loss.png" height="400px" width="400px" />
+</p>
+
+- #### **RESNET50**
+
+<p align="center">
+  <img src="https://github.com/Arihant-Bhandari/DL-Simplified/blob/fracture/Fracture%20Detection%20using%20DL/Images/RESNET50%20Accuracy.png" height="400px" width="400px" />
+  <img src="https://github.com/Arihant-Bhandari/DL-Simplified/blob/fracture/Fracture%20Detection%20using%20DL/Images/RESNET50%20Loss.png" height="400px" width="400px" />
+</p>
+
+- #### **Xception**
+
+<p align="center">
+  <img src="https://github.com/Arihant-Bhandari/DL-Simplified/blob/fracture/Fracture%20Detection%20using%20DL/Images/Xception%20Accuracy.png" height="400px" width="400px" />
+  <img src="https://github.com/Arihant-Bhandari/DL-Simplified/blob/fracture/Fracture%20Detection%20using%20DL/Images/Xception%20Loss.png" height="400px" width="400px" />
+</p>
+
+### 📈 **Performance of the Models based on the Accuracy Scores**
+
+#### Metrics: 
+
+We used  **Loss** and **Accuracy** as metrics.
+
+| Models | Validation Accuracy | Validation Loss | Test Accuracy | Test Loss |
+|--------|---------------------|--------------------------| ---------------------|--------------------------|
+| ResNet-50 | 38.74% | 9.8734 | 44.92% | 8.8777 |
+| InceptionV3 | 61.26%  | 6.1766 | 55.08% | 7.1615 |
+| CNN | 99.87% | 0.0100 | 100.00% | 0.0037 |
+| VGG16 | 94.63% | 0.1482 | 96.16% | 0.1142 |
+| MobileNet | 99.21% | 0.0346 | 100.00% | 0.0156 |
+| DenseNet-121 | 94.50% | 0.1413 | 95.03% | 0.1494 |
+| Xception | 98.82% | 0.0541 | 100.00% | 0.0170 |
+
+### 📢 **Conclusion**
+
+We conclude the following:
+
+**CNN**, **Xception**, **DenseNet-121**, **VGG16** and **MobileNet** are all up to the task and are ideal for this.
+
+### ✒️ **Your Signature**
+
+Original Contributor: Arihant Bhandari [https://github.com/Arihant-Bhandari]