Projects | Zakia Khatun

Advancing Human Soft Tissue Pathology Assessment Using Artificial Intelligence in Medical Imaging

Sat, 01 Feb 2025 00:00:00 +0000

Abstract

Artificial Intelligence (AI) has revolutionized various fields by automating complex tasks, uncovering patterns in large datasets, and making accurate predictions. Machine learning, particularly deep learning, plays a pivotal role in enabling AI to emulate human intelligence for tasks such as image segmentation, classification, and recognition. In the realm of medical imaging, modalities like Magnetic Resonance Imaging (MRI), Computed Tomography (CT), and Ultrasound are indispensable tools for diagnosing pathologies, detecting abnormalities, guiding treatment plans, and monitoring disease progression. The integration of AI with medical imaging offers the potential to enhance the accuracy and efficiency of these processes.

In particular, AI’s application in the assessment of tendon-related conditions, such as tendinopathy, presents a promising avenue for improving patient care. Tendinopathy can significantly impact a patient’s quality of life, and early detection is crucial to optimize treatment outcomes. This thesis focuses on the development of advanced AI-driven methods for analyzing human soft tissue pathologies, with a primary emphasis on tendon segmentation, pathology detection (classification), and tendon reflex response assessment. By automating the analysis of tendons and other human soft tissues, these methods aim to reduce human error and variability, thereby enabling more consistent and reliable clinical decisions. Ultimately, the goal is to support earlier, more accurate diagnoses and interventions, leading to better patient outcomes and more personalized treatment strategies.

This thesis begins with a study that analyzes MRI and CT scans from 47 participants to investigate the relationships between the tendons, cartilage, and muscles in the knee. This study has two primary objectives - first, to predict knee cartilage degeneration, and second, to predict patellar tendinopathy. For both objectives, predictions are made using features extracted solely from the patellar tendon and quadriceps, rather than directly from the cartilage itself. This approach explores the potential of using features from surrounding tissues as indirect predictors of knee-related pathologies. This study demonstrates that both knee cartilage degeneration and patellar tendinopathy can be predicted using these features from adjacent structures, highlighting the importance of surrounding tissues as potential indicators of pathology. Traditional machine learning models are employed to identify the most relevant features for each prediction task, highlighting their importance in the diagnosis of these conditions. This foundational research deepens our understanding of the interrelationships between knee soft tissues, contributing to more accurate diagnostic approaches in musculoskeletal health and enhancing clinical decision-making and treatment strategies.

A central focus of this thesis is the development of an end-to-end tendon segmentation module. This system integrates a superpixel-based coarse segmentation step that serves as a foundation for the final, more precise segmentation. In this approach, the segmentation task is framed as a superpixel classification problem. To achieve this, two distinct approaches are developed - (1) Random Forest (RF) and Support Vector Machine (SVM) classifiers for superpixel categorization, and (2) a Graph Convolutional Network (GCN) for transforming superpixels into graph structures for node classification. The RF and SVM classifiers demonstrate exceptional performance, achieving Area Under the Curve (AUC) scores of 0.992 and 0.987, respectively, with high sensitivity, indicating their effectiveness in accurately classifying superpixels. Although the GCN approach yields slightly lower performance, it showcases the potential of deep learning methods for improving segmentation by leveraging the structural relationships between superpixels. The findings suggest that both traditional machine learning and deep learning techniques offer promising avenues for advancing tendon segmentation, with superpixel-based methods offering a pathway to more reliable and automated segmentation in medical imaging.

Another key component of this thesis is the development of an end-to-end tendon pathology detection module, utilizing the same MRI dataset. This module adopts a graph-based approach, where superpixels are treated as nodes and connected by edge relationships. Each MRI scan is transformed into a graph, with the task framed as a graph classification problem to determine the presence or absence of pathology. To achieve this, a Graph Echo State Network (GESN) is employed. Known for its ability to efficiently represent data without the need for iterative backpropagation, the GESN leverages both temporal and structural dependencies in the data, enhancing classification performance. In this study, the GESN outperforms traditional machine learning models, achieving a mean accuracy of 0.953 and a sensitivity of 0.943. These results underscore the potential of the GESN to significantly enhance diagnostic accuracy, offering a powerful tool for early detection and clinical decision-making in tendon pathology assessment. Moreover, the GESN’s ability to handle complex, high-dimensional data suggests its broad applicability to other medical imaging tasks, further expanding its potential clinical utility.

The final study of this thesis explores the impact of demographic factors, including age, height, weight, and gender, on reflex response times in healthy individuals. This analysis is based on electromyography (EMG) recordings from 40 participants. The results reveal that elderly individuals, particularly those who are taller, heavier, and male, exhibit delayed reflex onsets. Even after normalizing for height, older participants still demonstrate slower reflex responses. These findings highlight the role of demographic factors in neuromuscular reflexes, aiding in the diagnosis and early detection of related disorders.

In conclusion, this research demonstrates the potential of AI, particularly superpixel-based and graph-based models, to advance tendon pathology assessment and exploratory tendon reflex studies, leading to better patient outcomes and musculoskeletal health management.

Survival Time Prediction of Metastatic Melanoma Patients by Computed Tomography using Convolutional Neural Networks

Sun, 30 Aug 2020 00:00:00 +0000

Metastatic melanoma is a fatal disease with a poor prognosis and rapid systemic spread. Follow-up and study are imperative, especially within the early periods after diagnosis as the expected cure is seldom obtained after surgical excision and with adjuvant therapy. Typically, computed tomography (CT) scan contains a huge sum of data that ought to be completely analyzed and assessed by the radiologist or other healthcare proficient in a brief time. In this case, the CAD framework can be of extraordinary offer assistance as a moment supposition for experts. In spite of the fact that the CAD framework may play an imperative part in the analysis, small research has been published on the survival time prediction of metastatic melanoma based on the CAD system.

Our objective is to study the prediction of the survival time of patients with metastatic melanoma in terms of 1-year survival as a binary classification. Dataset used in this study contains CTs of 71 patients with metastatic melanoma who are studied at Universite Clermont Auvergne Hospital. The number of lesions per patient varies from 1 to 11. To reach the objective, the survival time is anticipated using the accessible CT data as input of a 3D CNN. In this category, survival time is anticipated using full CT volumes and also from extracted 3D CT patches containing lesion regions. Here, the patches are extracted given the ground truth masks of the lesions. Moreover, segmentation of lesions coming from different organs is performed using two different 3D CNNs to examine the prediction of survival time based on newly extracted 3D CT patches. These new patches are extracted using our anticipated segmentation predicted masks of lesions. As the final test, it is also inspected whether aggregated deep segmentation feature map can help to predict survival time being an extra input channel to the CT data for 3D CNN or not.

Our study shows that using aggregated deep segmentation feature map as an extra input channel to CT data comes about in superior performance in survival time prediction compared to using as it were only 3D CT patches as input. In expansion, the prediction of survival time based on newly extracted 3D CT patches coming from our segmentation predicted masks is similar to the survival prediction using the 3D CT patches coming from ground truth masks.

Further investigation of this study can be the addition of radiomic features with aggregated deep segmentation feature map as additional input to CT data besides experimenting on bigger dataset. Including clinical data such as age, sex, etc. can as well play a vital role.

To Design, Analyze, and Implement Approaches for Brain Tissue Segmentation

Thu, 16 Jan 2020 00:00:00 +0000

The main goal of this project was to develop tissue (WM, GM, and CSF) segmentation methods in brain MRI images. The dataset used is IBSR18, which contains 18 skull-stripped and bias field-corrected T1-w images with different spatial resolutions (pixel spacing), and there is a heterogeneity in image intensities that hinders segmentation. System architecture used - Data → Pre-processing → Segmentation. Steps - 1) Pre-processing (a) Normalization and Skull stripping of MNI template, b) Image Registration (We have used Elastix as a software to apply 3D registration of moving image on fixed), c) Histogram Stretching (The registered volumes’ histograms have been stretched which broadens the histogram of the image intensity levels, d) Histogram Matching (Histogram stretched output of IBSR 07 has been taken as reference to match all other volumes’ stretched histogram), and 2) Segmentation - 3D patch-wise segmentation using 3D U-Net.

Developing Computer Aided Algorithm for the Diagnosis in Histopathological Images to Classify Benign vs Malignant Using Deep Learning Approach

Thu, 09 Jan 2020 00:00:00 +0000

The main objective of this project was to design a computer-aided diagnosis system to develop a deep learning algorithm for diagnosis in histological patches to classify benign vs malignant patches. System architecture used - Image → Normalization → Per instance standardization → Aggressive Data Augmentation → Classification. Steps - 1) Data normalization, 2) Per instance standardization (Standardization by mean and std of the training dataset), 3) Aggressive Data Augmentation (Horizontal flip, Vertical flip), and 4) Classification; Best performing classification model - ResnNet50 (ImageNet-pretrained).

Developing Computer Aided Algorithm for the Diagnosis in Dermoscopic Images to Classify Melanoma vs All Other Types Using Deep Learning Approach

Wed, 08 Jan 2020 00:00:00 +0000

The main goal of this project was to design a computer-aided diagnosis system to classify dermoscopic images and determine whether any case belongs to Nevus or lesion. System architecture - Image → Normalization → Per instance standardization → Aggressive Data Augmentation → Classification. Steps - 1) Data normalization, 2) Per instance standardization (Standardization by mean and std of the training dataset), 3) Aggressive Data Augmentation (Rotation, Width and Height shift, Zooming, Flipping, Brightness), 4) Classification; Best performing classification model - Ensemble of Inception ResNetv2, Inception v3, EfficientNet B3 (ImageNet-pretrained).

Image Registration of Chest CT Volumes; 4DCT DIR-Lab Challenge

Tue, 07 Jan 2020 00:00:00 +0000

This project aims to register 3D CT lung images, which was tested on the first 4 cases available on 4DCT DIR-Lab Challenge. Where data is available with 300 landmark annotations. Evaluation is made using TRE 3D Euclidean distance between transformed landmarks.

Demonstration of Medical Robotics in Spine Surgery using Tx60 Staubli Robot

Sun, 05 Jan 2020 00:00:00 +0000

The goal of this project was to demonstrate pedicle screw placement for spine fusion. Steps - 1) Trajectory planning (Type of movement is Point to point), 2) Software simulation of robotic spine surgery (Stäubli Robotics Suite), and 3) Testing demo surgery using Tx60 Stabuli.

Breast Mass Detection using Machine Learning Algorithms

Fri, 14 Jun 2019 00:00:00 +0000

The main aim of this project was to design a computer-aided diagnosis system that can detect mass/masses in mammograms to help the screening procedure. System architecture - Image → Candidate extraction → Feature extraction → Classification. Steps - 1) Candidate extraction- Histogram equalization → Multiple thresholding → Selection of n best threshold using mean and variance → Selection of desired threshold based on chosen threshold. 2) Feature extraction- Three types of features used are shape, texture (Haralick), and Intensity features. 3) Classification- Different type of classifiers were tested, both individually and as a cascade of two or more, to evaluate performance. At the end, the voting between Gaussian Naïve Bayes and Logistic regression was used where this voting is an AND combination of two classifiers.

CT Reconstruction using Parallelization Strategy

Tue, 11 Jun 2019 00:00:00 +0000

This project proposes a discrete image reconstruction algorithm using parallel beam geometry where a set of X-ray beams are passed through the object of interest and intensity variations of the beams at input and output are measured. To reconstruct, filtered back projection is used, which uses a 1D filter on the projection data before back projecting (2D or 3D) the data onto the image space.

Vision Application (VIZN)

Sun, 09 Jun 2019 00:00:00 +0000

The VIZN application provides an environment where users will be able to test some performance related to the eyesight, observing the evolution over time by means of innovative user controls. In this project, all the tests were prepared in a scientific way.

Designing an Inverse Kinematics Controller

Fri, 07 Jun 2019 00:00:00 +0000

The purpose of this project was to design an Inverse Kinematic controller for the Kinova Jaco2 robot to follow some given tasks.

Task 1 - The control objective is position only without exploiting the redundancy.

Task 2 - the control objective is given by both the position and the orientation. While the position needs to move according to the indications above, the orientation needs to be controlled such that it is kept constant at the initial value.

Task 3 - The end-effector orientation needs to be changed.

Task 4 - The redundancy needs to be exploited by maximizing the manipulability.

Interest Points Detection on 3D Meshes Using Harris Operator

Thu, 17 Jan 2019 00:00:00 +0000

In this project, an adaptive technique was implemented to determine interest points or 3D objects based on the Harris operator, which is done by determining the neighborhood of a vertex over which the Harris response is calculated. Steps - 1) For each vertex of mesh, forming a region of interest and it’s K neighborhood rings where all the rings are set of points, 2) Finding harris response, 3) Getting initial interest points, 4) Getting final interest points (either by selection fraction method or by clustering), 5) 3D rendering, and 6) Designing graphical user interface.

Automatic MRI Cardiac Segmentation in Short Axis for Left Ventricular Endocardium

Wed, 16 Jan 2019 00:00:00 +0000

The main aim of this project was to improve the accuracy of automatic Left ventricular (LV) segmentation in short-axis cardiac cine MR images. Steps - 1) Quantifying motion to determine an initial region of interest surrounding the heart, 2) Identifying potential 2D objects of interest using an intensity-based segmentation, 3)Assessing contraction/expansion, circularity, and proximity to lung tissue to score all objects of interest in terms of their likelihood of constituting part of the LV, and 4) Aggregating the objects into connected groups and construct the final LV blood pool volume and centroid.

Face Recognition using Principal Component Analysis

Tue, 15 Jan 2019 00:00:00 +0000

In this project, a simple face recognition system was designed based on a very small dataset of the training images. The two main steps of this project include data normalization and face recognition. After the normalization of the training data set, principle component analysis was incorporated. A simple Graphical User Interface was designed.

Automated Inspection at Soft Drink Bottling Plant

Sat, 12 Jan 2019 00:00:00 +0000

The main goal of this project was to develop a visual inspection system. The following seven fault conditions were aimed to be detected - Bottle under filled, Bottle over filled, Bottle has label missing, Bottle has label but label printing failed, Bottle label is not straight, Bottle cap is missing, and Bottle is deformed in some way. Background studies include extracting region of interest, computing mean gray level, binarizing any chosen region of interest with a particular thresholding, computing percentage of black pixels and comparing to normal bottle values, extracting red channel, edge detection and obtaining connected components & their bounding boxes, finding largest bounding box, comparing area, height & width to the normal. Depending on the fault condition, different types of steps were applied. A simple graphical user interface was designed.

Gesture Controlled Pick & Place Robot

Sat, 02 Dec 2017 00:00:00 +0000

In this project, a gesture-controlled pick-and-place robot was proposed with a drive system. This design is wirelessly controllable using a hand module. The main purpose was to aid physically disabled people to manipulate an object as they wish. Moreover, it will be useful in industrial work as it has the option of mobility, a trait that conventional pick-and-place robots do not have.