Learn More
Data Science Bootcamp
Commenced on 2–4 June 2025 at SNS, NUST. This website provides a comprehensive crux of all concepts, tools, and workflows demonstrated across the notebooks and visual material during the Bootcamp.
Foundational Data Science Tools
Python Fundamentals
Introduced Python syntax, data types, control structures, and functions for clean, modular code, forming the backbone for ML/DL pipelines.
NumPy & Pandas
NumPy covered array operations and mathematical computations. Pandas focused on data loading, cleaning, EDA, and preparing structured data for models.
Visualization & ML
Matplotlib/Seaborn for plotting data distributions and model performance. Scikit-learn introduced supervised learning (regression, trees, SVM).
Evaluation Metrics
Explained accuracy, precision, recall, F1-score, confusion matrix, and cross-validation, emphasizing their importance in clinical decision-making.
Image Processing and Neural Networks
Image Preprocessing
Covered image representation as pixel matrices, color spaces (RGB, grayscale), resizing, normalization, thresholding, blurring, and noise removal to prepare images for deep learning models.
Feature Extraction
Explored gradient-based methods like Sobel and Canny edge detection to extract low-level features such as edges and contours from images.
Dimensionality Reduction
Applied Principal Component Analysis (PCA) for feature compression and visualization of high-dimensional image data, preserving variance while reducing complexity.
Neural Network Basics
Explained artificial neurons, activation functions, loss functions, backpropagation, and optimization techniques to build intuition on how networks learn patterns.
CNN Architectures
Implemented Convolutional Neural Networks (CNNs) including convolution, pooling, and fully connected layers, demonstrating automatic learning of spatial features from images.
Edge Detection
Deep Learning in Medical Imaging
Medical Image Classification
Used eye image datasets to classify conditions like cataracts, covering data augmentation and CNN training in a medical context.
Transfer Learning
Applied pretrained VGG16/VGG19 models to medical tasks, discussing fine-tuning and the benefits for small datasets.
MRI and CT Analysis
Worked with MRI/CT scans for disease classification, focusing on domain-specific preprocessing and handling challenges like noise and class imbalance.
Chest Scan (Pneumonia)
Advanced Deep Learning Techniques
Image Segmentation (U-Net)
Implemented U-Net architecture for pixel-wise segmentation, using encoder-decoder structure and skip connections to identify regions of interest like tumors.
Object Detection Fundamentals
Introduced concepts like bounding boxes, IoU, and confidence scores, differentiating between classification, localization, and detection tasks.
YOLO for Detection
Applied YOLO models for real-time object detection on medical images, including brain tumor localization, covering annotation and inference.
Model Training & Inference
Demonstrated model checkpointing, saving trained weights, loading models for inference, and testing on unseen data for deployment readiness.
Generative Models and Ethical AI
Generative Models
Introduced Variational Autoencoders (VAEs) for learning latent representations and Diffusion Models for high-quality image synthesis, enhancement, and augmentation.
Ethics and Deployment
Discussed ethical considerations in medical AI, dataset bias, model limitations, explainability, and the importance of human oversight for responsible AI usage.
Comprehensive Learning Journey
The bootcamp covered data science foundations, ML, DL, and advanced medical imaging applications, demonstrating educational implementations of classification, segmentation, detection, and generative modeling.
From Python to Pandas, NUMPY, Matplotlib, Scikitlearn, Machine learning, deep learning, image processing, GANS, Yolo.
Visualizing Edge Detection and Data Components
Canny object detection results demonstrate feature extraction techniques.
Example of the Eye dataset used for medical image classification.
Visualization of PCA Components.
Preparing Dataset for Deep Learning (Tumour Detection)
Model Performance Visualization
Visual representations of model training and evaluation metrics.
U-Net Architecture Implementation
The code snippet below outlines the U-Net model construction using Keras layers for medical image segmentation. It includes definitions for the convolutional block, encoder block (with pooling), and decoder block (with skip connections and transpose convolution).
from tensorflow.keras.layers import Input, Conv2D, Conv2DTranspose, MaxPooling2D, Concatenate
from tensorflow.keras.models import Model

# conv block
def conv_block(input, num_filters):
    conv = Conv2D(num_filters, (3, 3), activation="relu", padding="same", kernel_initializer='he_normal')(input)
    conv = Conv2D(num_filters, (3, 3), activation="relu", padding="same", kernel_initializer='he_normal')(conv)
    return conv

# Encoder block
def encoder_block(input, num_filters):
    conv = conv_block(input, num_filters)
    pool = MaxPooling2D((2, 2))(conv)
    return conv, pool

# Decoder block
def decoder_block(input, skip_features, num_filters):
    uconv = Conv2DTranspose(num_filters, (2, 2), strides=2, padding="same")(input)
    con = Concatenate()([uconv, skip_features])
    conv = conv_block(con, num_filters)
    return conv

# Build model
def build_model(input_shape):
    input_layer = Input(input_shape)
    s1, p1 = encoder_block(input_layer, 64)
    s2, p2 = encoder_block(p1, 128)
    s3, p3 = encoder_block(p2, 256)
    s4, p4 = encoder_block(p3, 512)

    b1 = conv_block(p4, 1024)

    d1 = decoder_block(b1, s4, 512)
    d2 = decoder_block(d1, s3, 256)
    d3 = decoder_block(d2, s2, 128)
    d4 = decoder_block(d3, s1, 64)

    output_layer = Conv2D(1, 1, padding="same", activation="sigmoid")(d4)
    model = Model(inputs=input_layer, outputs=output_layer, name="U-Net")
    return model
The U-Net model is built and compiled for binary classification tasks.
Brain Tumor Detection with YOLO
Object detection using YOLO was applied for tasks like brain tumor localization.
Generative Modeling Visuals
Variational Autoencoders (VAEs)
VAEs are used for learning latent data representations, useful for anomaly detection and data generation.
Diffusion Models
These models generate high-quality images by progressively removing noise, relevant for medical image synthesis and enhancement.
Made with