Tajwar Abrar Aleef

Hi! I am a PhD Candidate from the Robotics and Control Laboratory at the University of British Columbia (Vancouver, Canada). I am working with Vancouver General Hospital & BC Cancer (Vancouver, Canada) to improve ultrasound-guided prostate cancer detection & treatment. One of my current projects is on developing a multi-parametric ultrasound system that can image the mechanical properties of tissue and aid in the non-invasive detection of prostate cancer.

I am an Erasmus Scholar and my MSc was a specialized degree focused on Medical Imaging fully funded by the European Union and was jointly hosted by three European countries (France, Italy, and Spain). Additionally, I hold a BSc degree in Electrical & Electronic Engineering. Overall, I have a strong background in AI & Medical Imaging and have many years of experience working with multiple research groups & hospitals across Europe and Canada.

Outside research, I play guitar and enjoy all sorts of sports (nowadays lifting, climbing, and ice-skating). I also enjoy traveling a lot; so far have covered 23 countries and countless cities (here's a vlog from Thailand).

Key research interests: Medical Imaging & Analysis, Artificial Intelligence, Image Processing, Computer Vision, Ultrasound Imaging, Ultrasound Elastography.

Email  |  CV  |  Google Scholar  |  GitHub  |  Research Gate  |  LinkedIn  |  Twitter  |  YouTube

profile photo
Selected Research Works

Most of my research work is centered around cancer/object of interest detection/localization/segmentation, solving constrained optimization problems for radiotherapy, adversarial learning, ultrasound elastography, and designing biomedical implantable antennas.

3D Shear Wave Absolute Vibro-Elastography for guided prostate biopsy: phantom and in vivo liver validation
Tajwar Abrar Aleef, Qi Zeng, Hamid Moradi, Shahed Mohammed, Tom Curran, Mohammad Honarvar, W. James Morris, S. Sara Mahdavi, Septimiu E. Salcudean
Submitted to IEEE IUS, 2022  

System with an end-firing probe mounted shaker for doing shear wave absolute vibro-elastography of prostate. The system has been validated by comparing with MRE results using quality assurance phantoms and in vivo liver data.
Quasi-Real Time Multi-Frequency 3D Shear Wave Absolute Vibro-Elastography (S-WAVE) System for Prostate
Tajwar Abrar Aleef, Julio Lobo, Ali Baghani, Hani Eskandari, Hamid Moradi, Robert Rohling, S. Larry Goldenberg, William James Morris, S. Sara Mahdavi, Septimiu E. Salcudean
Submitted to IEEE TMI, 2022  
[ arXiv ]

A system developed for 3D elastography for prostate using a transperineal exciter. The system is validated using quality assurance phantoms and two clinical patient studies.
Registration of transperineal template mapping biopsy cores to volumetric ultrasound
Tajwar Abrar Aleef, Qi Zeng, W. James Morris, S. Sara Mahdavi, Septimiu E. Salcudean
IPCAI, 2022 (Oral Presentation to be held in Tokyo)  
[ Paper  |  Code ]

This framework uses information taken from the existing workflow of transperineal template mapping biopsy (such as the needle insertion point and depth) to register the pathology of the cores to pre-operative ultrasound volumes. Pathology of the cores will be used as labels to train prostate cancer classifiers using multi-parametric ultrasound volumes.
Rapid treatment planning for low-dose-rate prostate brachytherapy with TP-GAN
Tajwar Abrar Aleef, Ingrid T. Spadinger, Michael D. Peacock, Septimiu E. Salcudean, S. Sara Mahdavi
MICCAI, 2021 (Oral Presentation)  
[ Interview  |  Paper  |  arXiv  |  Code ]

An end-to-end framework capable of generating treatment plans for low-dose-rate prostate brachytherapy within seconds compared to manually creating plans (current standard) which takes over 20 min. This is done utilizing GANs with custom application-specific loss functions.
Centre-specific autonomous treatment plans for prostate brachytherapy using cGANs
Tajwar Abrar Aleef, Ingrid T. Spadinger, Michael D. Peacock, Septimiu E. Salcudean, S. Sara Mahdavi
IPCAI, 2021   (Oral Presentation, Honorary Award for Best Presentation)
[ Paper  |  Video  |  Code ]

A novel approach to automating centre-specific treatment plans for low-dose-rate prostate brachytherapy. Conditional GAN was used to solve the optimization problem of placing needles into the prostate gland. These needles allow seeds to be placed in a way that radiates the gland alone and spares healthy surrounding tissues.
Co-Generation and Segmentation for Generalized Surgical Instrument Segmentation on Unlabelled Data
Megha Kalia**, Tajwar Abrar Aleef**, Nassir Navab, Peter Black, Septimiu E. Salcudean
MICCAI, 2021 (** joint first author) (Oral Presentation, MICCAI travel award)  
[ Paper  |  arXiv  |  Code ]

We propose a novel joint generation and segmentation strategy to learn a segmentation model with generalization capability to domains containing no labeled data. Such a technique is crucial in this application as there exists only limited labeled surgical robot-assisted prostatectomy data.
Thyroid Nodule Segmentation
Tajwar Abrar Aleef, Hasib Zunair
MICCAI TN-SCUI Challenge, 2020

In this work, several state-of-the-art image segmentation techniques were explored for optimizing Thyroid Nodule segmentation from Ultrasound images. This inlcuded supervised learning, transfer learning, and generative adversarial learning.
Automatic Mass Classification in Breast Using Transfer Learning of Deep Convolutional Neural Network and Support Vector Machine
Md. Kamrul Hasan, Tajwar Abrar Aleef, Shidhartho Roy
IEEE Region 10 Symposium (TENSYMP), 2020 (Oral Presentation)
[ Paper ]

In this work, an automated classifier was designed for the classification of mass from mammograms. The proposed pipeline uses DCNN for feature extraction, bagged decision tree for feature selection, and an SVM classifier for distinguishing between tissue with/without mass.
Malignancy estimation of pulmonary nodules using multi-view multi-time point convolutional neural networks
Tajwar Abrar Aleef, Colin Jacobs, Bram van Ginneken
MSc Thesis, DIAG, Radboudumc, 2018   (Ranked Top 2nd)

A novel multi-view multi-timepoint convolutional neural network (MVMT-CNN) was proposed to estimate malignancy in pulmonary lung nodules. We utilize scans of patients taken over time- providing the network with information on disease progression. This technique produced state-of-the-art results compared to standard single time-point disease estimation.
Design, Measurement and Performance Enhancement of a Miniaturized Implantable Antenna for Biomedical Applications
Tajwar Abrar Aleef, Yeman Brhane Hagos, Vu Hoang Minh, Saed Khawaldeh, Usama Pervaiz,
Micro & Nano Letters, 2017
[ Paper ]

A biomedical antenna was designed for wireless communication from inside a living body. The operating band for this antenna is in the Industrial, Scientific, and Medical band (2.4–2.48 GHz). The tiny dimension of the antenna with the 9.45 μm thickness of the patch allows it to be highly flexible- providing excellent results even in extreme conditions where the antenna conforms along the curvature of the body.
CNN regressor for automating treatment planning of cervical brachytherapy
Tajwar Abrar Aleef, Marius Staring, Jan van Gemert
LKEB, LUMC, 2017
[ Report  |  Code ]

Developed a framework using CNNs that can predict the geometric transformation of an object based on a target. The idea was to analyze if such a regression network can be used to find the transformation parameters of a brachytherapy applicator in the cervix from MRI scans; hence pinpointing the exact location of the device within the body.
Fast pet scan tumor segmentation using superpixels, principal component analysis and k-means clustering
Yeman Brhane Hagos, Vu Hoang Minh, Saed Khawaldeh, Usama Pervaiz, Tajwar Abrar Aleef
Methods and Protocols, 2018
[ Paper ]

This work presents a fast positron emission tomography tumor segmentation method using superpixels, principal component analysis, and k-means clustering.
Complete End-To-End Low Cost Solution To a 3D Scanning System with Integrated Turntable
Saed Khawaldeh, Tajwar Abrar Aleef, Usama Pervaiz, Yeman Brhane Hagos, Vu Hoang Minh
[ arXiv ]

In this project, a full 3D scanning system was designed along with a low-cost turntable. The software included a GUI and was implemented with C++ and Point cloud Library (PCL) that uses a Kinect sensor as the sensor. The point clouds are then passed through a filter followed by registration and reconstruction to generate the 3D data.
Skin Cancer Diagnosis Tool
Tajwar Abrar Aleef, Yuliia Kamkova
[ Code ]

This UI lets users register an account along with their Doctor's information. The user can then either take new images or load images from the disk and the toolkit will run the diagnosis in the backend (based on pretrained AlexNet) and send the results directly to the doctor. The doctor makes the final diagnosis and consults the user.
Study of Macular Lesion classification and Retinal vessel segmentation using deep learning techniques
Tajwar Abrar Aleef, Yuliia Kamkova

In this work, first, a classification problem was tackled for classifying macular patches into three classes. Several deep learning approaches were studied including pre-trained models such as the AlexNet, VGG, GoogleNet, and resNet50. For Retinal vessel segmentation, a patch-based segmentation approach was used, where, a simple CNN architecture was trained using 120,000 patches taken from the training set.
Heart flow analysis in ECG-gated cardiac Ultrasound
Tajwar Abrar Aleef, Yeman Brhane Hagos, Vu Hoang Minh

Automatic spatio-temporal analysis of cardiac flow was performed using Lucas Kanade and Farneback optical flow techniques. This method tracks the ventricle wall and displays sparse and dense patterns of the vector field of the blood flow. Local and global maximum velocities of the flow are also calculated and displayed for each frame. If any vorticity is detected, it is displayed using colormaps.
Past Engineering Projects

I always enjoyed hands-on work. Even long before I got any formal education or training on engineering, I was often found meddling around with electronic devices I could get my hands on. The goal was to take them apart, learn how they operate, and "hopefully" make something new out of it. Although I started doing this from a very young age, I only started documenting when my parents got me a digital camera. I have added some of these documented builds below (most of them are from my high school and freshman year). Scroll to the bottom to see some janky yet wonderful rovers I made literally out of cardboard and salvaged parts from old toys.

5-Axis Robotic Arm
[ Video ]

Made this totally from scratch and with no access to proper tools. The parts were scorched and carved out of plastic sheets with just a box cutter. Rubber from a balloon was stapled onto the gripper for extra grip (my mom's idea). The 6 servos at each articulation point are controlled by an Arduino Mega board. The program I wrote for this uses inverse kinematics to control the position of the end effector. In the hyperlinked video, the arm is programmed to move to position A, grab the object, take it to position B, rotate it, and drop it.
Firefighting Bot
[ Video ]

Concept of autonomous fire detection and extinguishing. The bot has IR sensors mounted in front that search for fire in the surrounding environment. If a fire is detected, it goes towards it and turns on the extinguisher (for simplicity a fan is used). Once the fire is out, the robot backs up and starts looking around for more possible fires.
Surveillance Robot
[ Video ]

Here, I made a custom android controller allowing teleoperation of the robot over Bluetooth. Another Android phone mounted on the robot acts as an IP camera and sends a live video feed to the operator. The robot can also automatically avoid obstacles using its onboard sonar sensor.
Sonar Glasses for the visually impaired
[ Video ]

This is a simple concept of using sonar sensors to assist visually impaired people. These sensors were mounted onto glasses that detect if any obstacle is present in the periphery. Based on the position of the object, two vibration motors mounted on each side of the glasses will vibrate with different amplitude- giving haptic feedback to the user and alerting them of nearby objects.
Smart Line Follower
[ Video (contains unprofessional jokes!)]

In my freshman year, as I started learning more about logic gates. I decided to make a line follower out of just logic gates. Happy with the results, I wanted to see how much is possible with logic gates alone and with no microcontroller. The robot has 1 bit of memory made from 1 JK flip-flop (also made out of logic gates) which stores the state of the robot. If the robot detects a payload, it will start its journey to the other side of the line. And only after the payload is removed, the robot will return. Check the ridiculous (and a bit inappropriate) story I made with this robot in the video.
[ Video ]

Check the rovers below before reading this. Still obsessed with rovers, this was another attempt to make further improvements. This version again got a gripper redesign. In this build, I figured out how to melt and shape empty pen casings to make a stronger gripper as opposed to the popsicle stick technique. The arm was pushed back into the rover- allowing it to lift even heavier loads. Other features I added include an IP camera (using a phone) to stream live video which can also be rotated giving 360 degrees vision during teleoperation. During that time, I also got a fancy camera which I used to film the rover. Check the video where I made a short story using the rover as the hero. Interesting fact: Around that time, I just started to learn about circuits from my A-level physics course. I implemented a circuit in this rover that automatically controls the headlights based on the ambient light and thought it was super cool.
[ Video  |  Short Video ]

This is the updated version of RRL1. After finding success with the electronic side of things, in this build, I focused on improving the mechanical design. Again with no proper tools and materials available, I used things I found around the house. The gripper design was improved substantially and now it could pick up many different items. In the 11 min long video, I show the things I could do with this version (even tried to pick up a pen and draw with it). Interesting fact: I didn't have any servos back then so I made a string pulley system to increase the load capacity of the DC motor that raises the arm.
[ Video ]

At the age of 14, I became obsessed with rovers after seeing a documentary on NASA's Mars rovers: Spirit and Opportunity. Following that day, I have spent countless hours in my room trying to learn things by myself with the hopes of building my own rover one day. After countless failed iterations, this was the first working prototype. If you observe closely, this was made mostly out of cardboard, popsicle sticks, and foam board. The electronic parts were taken from RC cars. Fun fact: I was so happy with how it turned out that I named it after my high school sweetheart.
Teaching Experience

Based on Jon Barron's site.
Last updated December 2021.