Oral Disease Detection using Deep Learning and Computer Vision

Deep Learning, Computer Vision

Deep Convolutional Neural Networks were trained with over a thousand instances of annotated panoramic dental X-rays to identify teeth which contained Caries within them.

Our software suite consists of Dentaid Server, Dentaid OPG Client and Dentaid Diagnostics (a web portal for dentists to view diagnosed OPGs and diagnose their own X-rays) which can be setup in any hospital. Our software reads data from the radiologists x-ray machine real time and sends it to the cloud where it is diagnosed and is then sent to the dentists' web portal where they can view the annotated OPG with dental caries detected along with the patient name and other records.

The neural network was built on Keras with TensorFlow backend and trained on Google Collab, Dentaid Server was built as a REST API server on Flask, deployed on Microsoft Azure while we used Google Firebase for our storage bucket and database and our front end application was built on Angular.

Data acquisition, annotation and pre processing were part of the process.

This software suite was developed as part of our Final Year Project and in collaboration with DOW University of Health Sciences, Ojha Campus.

Identify Customer Segments

Unsupervised Learning, Principle Component Analysis, K-Means Algorithm

Unsupervised learning techniques were applied to identify segments of the population that form the core customer base for a mail-order sales company in Germany.

These segments can then be used to direct marketing campaigns towards audiences that will have the highest expected rate of returns.

The data that will be used has been provided by Bertelsmann Arvato Analytics, and represents a real-life data science task.

Finding Donors for Charity

Supervised Learning,Random-forest, Naive-bayes-classifier, Logistic-regression

Project revolved around building an algorithm for a fictitious charity organization to best identify potential donors and reduce overhead cost of sending mail.

The goal was to evaluate and optimize several different supervised learners to determine which ML algorithm will provide the highest donation yield while also reducing the total number of letters being sent.

Explore US Bikeshare data

Python, Data Analysis

Using data provided by Motivate, a bike share system provider for many major cities in the United States, this command line application uncovers bike share usage patterns in the three major US cities: Chicago, New York City, and Washington, DC.

This program is easily downloadable and executable via python command line.

ETL & Data Visualization

SSAS, PowerBI, SSIS

Performed incremental data refresh (UPSERT) on an original dataset using SSAS and SSIS. Used Tableau and Power BI for data visualization.

Image Classifier

Deep Learning, Transfer Learning

An image classification application which identifies and predicts flowers according to species using a deep neural network which has been trained using a VGG16 Model using transfer learning. The prediction accuracy is 86%.

This application can be run from command line and allows the user to train the model according to various specifications e.g. VGG16, ResNET etc., Architecture type etc.

Investigate IMDB dataset

Data-visualization, Data-analysis, Seaborn

IMDb, the Internet Movie Database, has been a popular source for data analysis and visualizations over the years. The combination of user ratings for movies and detailed movie metadata have always been fun to play with.

Model United Nations - Delegate Registration Portal

Angular4, noSQL, Firebase, Reactive Forms

A Delegate Registration portal for Model United Nations at IBA Karachi(MUNIK) with interactive front end, back end and database using Angular 4, Bootstrap and Firebase.

Application allowed dynamic user attribute modification allowing the administrator to add/remove/modify delegate/user attributes such as committee name, country allocated, delegation verified. Angular 4 reactive forms were extensively used.

This was my first commercial application project created in collaboration with Omer Farooq Ahmed.

Careem/Uber online portal

Bootstrap 4, Coursera Web-Application Specialization

Created a Front-End Web-Application (User Interface only) as a Coursera capstone project, using Bootstrap 4 for car ride services such as Uber and Careem.

Website Development

CLICK TO VIEW WEBSITES

Contact Book Application

Data Structures and Searching Algorithms

Created a contacts manager application with fast search features using Trie data structure.

Visual Design using PROCESSING IDE

Developed basic games i.e. Ludo, Jigsaw Match etc. using Processing IDE. Also created Lindenmayer system to convert strings into geometric patterns.

Graphic Design and Event Branding