Embedded Files
This research work received 2 Best Paper Awards at the SPIE, Computer-Aided Diagnostics (CAD) Conference.
Brain Tumor Recurrence vs. Radiation Necrosis Classification and Patient Survivability Prediction
Brain Tumor Recurrence vs. Radiation Necrosis Classification and Patient Survivability Prediction
Glioblastoma (GB) is a highly aggressive brain tumor with poor survival rates despite intensive treatments. Differentiating between radiation-induced necrosis (RN) and recurrent brain tumor (rBT) on MRI is crucial for timely intervention but challenging due to their similar imaging characteristics. This study addresses the challenge of class imbalance in MRI-based classification by employing a novel computational approach with multiresolution radiomic features (MRF) and statistically rigorous repeated random sub-sampling. The model demonstrated high performance in classifying rBT from RN, achieving an area under the curve (AUC) of 0.892±0.055. Additionally, the MRF model effectively predicted patient survival, with an AUC of 0.77±0.032. This approach not only improves classification accuracy but also offers a non-invasive biomarker for identifying patients at higher risk of recurrence or radiation necrosis, outperforming current state-of-the-art methods.
*Image copyright will be transferred to IEEE Journal of Biomedical & Health Informatics
Uncertainty Estimation in Classification of MGMT Using Radiogenomics for Glioblastoma Patients
Uncertainty Estimation in Classification of MGMT Using Radiogenomics for Glioblastoma Patients
Glioblastoma Multiforme (GBM) is one of the most malignant brain tumors among all high-grade brain cancers. Temozolomide (TMZ) is the first-line chemotherapeutic regimen for glioblastoma patients. The methylation status of the O6-methylguanine-DNA-methyltransferase (MGMT) gene is a prognostic biomarker for tumor sensitivity to TMZ chemotherapy. However, the standardized procedure for assessing the methylation status of MGMT is an invasive surgical biopsy, and accuracy is susceptible to resection sample and heterogeneity of the tumor. Recently, radio-genomics which associates radiological image phenotype with genetic or molecular mutations has shown promise in the non-invasive assessment of radiotherapeutic treatment. This study proposes a machine-learning framework for MGMT classification with uncertainty analysis utilizing imaging features extracted from multimodal magnetic resonance imaging (mMRI). The imaging features include conventional texture,volumetric, and sophisticated fractal, and multi-resolution fractal texture features. The proposed method is evaluated with publicly available BraTS-TCIA-GBM pre-operative scans and TCGA datasets with 114 patients.The experiment with 10-fold cross-validation suggests that the fractal and multi-resolution fractal texture featuresoffer an improved prediction of MGMT status. The uncertainty analysis using an ensemble of StochasticGradient Langevin Boosting models along with multi-resolution fractal features offers an accuracy of 71.74% andarea under the curve of 0.76. Finally, analysis shows that our proposed method with uncertainty analysis offers improved predictive performance when compared with different well-known methods in the literature.
Obesity Prediction from Structural MRI Using Conformal Deep Learning with Uncertainty Quantification
Obesity Prediction from Structural MRI Using Conformal Deep Learning with Uncertainty Quantification
Obesity increases the risk of brain diseases such as dementia and cognitive decline, which lead to neurobiological alterations. One key phenotype associated with obesity is body mass index (BMI). BMI is influenced by multiple causal pathways driven by behavioral, metabolic, and genetic factors. Traditional obesity prediction studies often rely on magnetic resonance imaging (MRI) voxel-based morphometry to correlate BMI with obesity-related clinical measurements and brain structure, predominantly grey matter volume (GMV). However, these studies often show inconsistencies in regional GMV findings among obese patients, making it challenging to establish a clear relationship between obesity and brain structure. To address these limitations, we propose a computational model to predict obesity directly from individual T1-weighted structural MRI data. The proposed conformal deep learning model achieves a 5-fold cross validated average precision of 77.65% and an F1-score of 75.42%, effectively predicting the probabilistic outcome of obesity from structural MRI. Furthermore, our model provides probabilistic uncertainty quantification paired with gradient-based localization maps that discover key brainregions, such as lobes and white matter tracts for obesity prediction. Comparisons with standard segmented brain atlas and existing literature demonstrate that our CDL model offers complementary evidence linking obesity with structural MRI regions. These experimental results present a promising approach for further investigation of clinical utility of MRI-based obesity prediction.
Domain Adaptive Federated Learning for Multi institution Molecular Mutation Prediction and Bias Identification
Domain Adaptive Federated Learning for Multi institution Molecular Mutation Prediction and Bias Identification
Deep learning models have shown potential in medical image analysis tasks. However, training a generalized deep learning model requires huge amounts of patient data that is usually gathered from multiple institutions which may raise privacy concerns. Federated learning (FL) provides an alternative to sharing data across institutions. Nonetheless, FL is susceptible to a few challenges including inversion attacks on model weights, heterogenous data distributions, and bias. This study addresses heterogeneity and bias issues for multi-institution patient data by proposing domain adaptive FL modeling using several radiomics (volume, fractal, texture) features for O6-methylguanine-DNA methyltransferase (MGMT) classification across multiple institutions. The proposed domain adaptive FL MGMT classification inherently offers differential privacy (DP) for the patient data. For domain adaptation two techniques e.g., mixture of experts (ME) with a gating network and adversarial alignment are used for comparison. The proposed method is evaluated using publicly available multi-institution(UPENN-GBM, UCSF-PDGM, RSNA-ASNR-MICCAI BraTS-2021) data set with a total of 1007 patients. Our experiments with 5-fold cross validation suggest that domain adaptive FL offers improved performance with a mean accuracy of 69.93% ± 4.8 % and area under curve of 0.655 ± 0.055 across multiple institutions. In addition,further analysis of probability density of gating network for domain adaptive FL identifies the institution that may bias the global model prediction due to increased heterogeneity for a given input. Our comparison analysis shows that the proposed method with bias identification offers the best predictive performance when compared to different commonly employed FL and baseline methods in the literature
"Bolte Chai Plus"- an Intelligent Integrated Platform
"Bolte Chai Plus"- an Intelligent Integrated Platform
Functional communication is indispensable for child development but worldwide children with autism and neurodevelopment disorder (NDD) combats communication barrier even though they are aspired to communicate their needs and desires. In Bangladesh, there are nearly half million verbally challenged children with autism and NDD who face challenges in initiating day-to-day communication and acquiring basic education. To empower the voice of these children and to ensure their rudimentary rights, we are proposing “Bolte Chai-Plus” platform, which encompasses Android and iOS mobile application, integrated website and intelligent analytics. With the mobile application children will be able to express their desire through voice generated from smartphone when activity picture is selected. Additionally, with help option children can seek assistance by sending SMS to parents by selecting their picture. Moreover, parents, caregiver, teacher, Speech Language Therapist can personalize activities and track children learning progress via individual mobile application account or website by utilizing unique user profile. Furthermore, the platform ensures institutional benefit where institutions associated with special education located at different corner of Bangladesh come under one single platform which enables tracking of user number, user management and remote monitoring of autism and NDD children communication development.
Undergraduate Thesis Title: "Optimization of features for Classification of Parkinson’s Disease from Vocal Dysphonia"
Undergraduate Thesis Title: "Optimization of features for Classification of Parkinson’s Disease from Vocal Dysphonia"
Parkinson’s Disease is considered most prominent neurological disease after Alzheimer and Epilepsy. There is no defined test for early diagnosis of Parkinson’s Patient and medical decisions are provided based on the medical history of the patient and hence the possibility of misdiagnosis. Parkinson’s disease influxes different prospects of a patient such as Gait, EEG and Speech and in 90 percent cases vocal dysphonia is present and analysis of the vocal dysphonia can be considered as the early biomarker of decision making for medical practitioners and neurologists as well as biometric analysis. This study aims at vocal dysphonia analysis of Parkinson’s Patient from voice dataset with different machine learning algorithms using WEKA and Python Programming. The objective is comparison of performance evaluation parameters of machine learning algorithms with previous work and proposing relevant features related to respective algorithms methods and more efficient classification using machine learning algorithms such as: Support Vector Machine (SVM), k-Nearest Neighbors , Random Tree, Naïve Bayes ,Logistic Regression and Multi-Layer Perceptions. A comparative study is performed between WEKA and Python Programming where k-Nearest performed approximately same 98% accuracy with 5 relevant attributes. In addition in case of Multi-Layer Perceptions with different hidden layers the performance is evaluated. It is observed that MDVP:Fo, MDVP:Shimmer, RPDE, Spread1 attributes contributes more for efficient classification accuracy which is intertwined with previous work on the same dataset and in most cases results in better performance.
Page updated
Google Sites
Report abuse