Tuberculosis (TB) Detection

From Chest X-ray

Fahim Shahriar (Anik)- 1605065

Shovito Barua Soumma - 1605066

Info

Architecture – CNN

Dataset – Kaggle (3500 Normal Images and 700 TB images)

(https://www.kaggle.com/tawsifurrahman/tuberculosis-tb-chest-xray-dataset)

Framework- Tensorflow, Keras

Performance measure: Accuracy

​

Relevant Paper: Reliable Tuberculosis Detection Using Chest X-Ray With Deep Learning, Segmentation and Visualization

​

​

Normal

TB

Preprocessing Steps

• By default: 512 X 512 pixels
• Need to resize the shape according to Model:
• InceptionV3 : 300 × 300 pixels
• ChexNet, MobileNetV2: 224 × 224 pixels
• Custom NN: 64 x 64 pixels (2 CNN, 3 Dense Layer)
• Split the dataset into train, test and validation

Preprocessing Steps

Preprocessing Steps

• Split the dataset into train, test and validation
• Augmentation: Training set

​

Custom Model

Custom Model

63

​

Transfer Learning (InceptionV3)

Transfer Learning (InceptionV3)

64

Transfer Learning (ChexNet)

Transfer Learning (ChexNet)

56

Transfer Learning (MobileNetV2)

Transfer Learning (MobileNetV2)

Epoch = 15

Transfer Learning (MobileNetV2)

Epoch = 15

Transfer Learning (MobileNetV2)

Epoch = 15

Transfer Learning (MobileNetV2)

Epoch = 5

Transfer Learning (MobileNetV2)

Epoch = 15

57

Challenges

• Imbalanced dataset
• Lack of data
• Overfitting issue

​

​

Thank You