Shahabuddin Siddiqui

Neurological disorders are among the leading causes of disability and mortality worldwide, making timely and accurate diagnosis essential. Differential diagnosis (DD) of Multiple Sclerosis (MS) and Neuromyelitis Optica (NMO) spectrum disorder using brain MRI remains challenging due to overlapping clinical manifestations and similar lesion characteristics. To address this challenge, this study proposes a neuroinformatics-driven computer-aided differential diagnosis (CADD) framework that automatically extracts, encodes, and analyzes quantitative 3D lesion biomarkers from brain MRI using deep learning and interpretable neuroimaging analytics. The proposed framework integrates three artificial intelligence (AI) modules: the NeuroImaging Sequence Examiner (NISE) for MRI sequence identification, the NeuroImaging Volumetric Extractor (NIVE) for skull stripping, and the NeuroImaging Lesion Extractor (NILE) for lesion segmentation. These modules are combined into an end-to-end neuroinformatics framework termed NeuroImaging Differential Evaluation (NIDE), which performs automated neuroimaging processing, lesion biomarker extraction, structured feature encoding, and differential diagnosis. Following lesion segmentation, quantitative 3D morphological and geometric lesion properties are extracted and transformed into a structured neuroimaging representation, where each row corresponds to an individual lesion and each column encodes a specific lesion biomarker. The resulting feature matrix is subsequently classified using a 2D convolutional neural network (2D-CNN). Comparative experiments against 3D-CNN models trained on raw MRI volumes, skull-stripped MRI volumes, lesion-mask volumes, and conventional machine learning classifiers demonstrate the superiority of the proposed lesion-driven learning strategy. Using ResNet50, the proposed framework achieved an accuracy of 80% for MS–NMO spectrum disorder differential diagnosis. Furthermore, the framework is independent of MRI view-plane orientation, enabling robust performance across different imaging orientations while operating in a fully automated manner. Beyond classification performance, the framework provides a structured representation of quantitative lesion biomarkers and enables interpretable neuroimaging analytics through Explainable Artificial Intelligence (XAI), with model interpretations corroborated by expert radiological assessment. The proposed NIDE framework demonstrates how automated neuroimaging processing, quantitative lesion characterization, structured feature encoding, and explainable AI can be integrated into a unified neuroinformatics workflow for differential diagnosis. These findings highlight the potential of neuroinformatics-driven decision-support systems for advancing neuroimaging analysis and clinical management of demyelinating neurological disorders.

Medical imaging plays a pivotal role in diagnostic medicine with technologies like Magnetic Resonance Imagining (MRI), Computed Tomography (CT), Positron Emission Tomography (PET), and ultrasound scans being widely used to assist radiologists and medical experts in reaching concrete diagnosis. Given the recent massive uplift in the storage and processing capabilities of computers, and the publicly available big data, Artificial Intelligence (AI) has also started contributing to improving diagnostic radiology. Edge computing devices and handheld gadgets can serve as useful tools to process medical data in remote areas with limited network and computational resources. In this research, the capabilities of multiple platforms are evaluated for the real-time deployment of diagnostic tools. MRI classification and segmentation applications developed in previous studies are used for testing the performance using different hardware and software configurations. Cost–benefit analysis is carried out using a workstation with a NVIDIA Graphics Processing Unit (GPU), Jetson Xavier NX, Raspberry Pi 4B, and Android phone, using MATLAB, Python, and Android Studio. The mean computational times for the classification app on the PC, Jetson Xavier NX, and Raspberry Pi are 1.2074, 3.7627, and 3.4747 s, respectively. On the low-cost Android phone, this time is observed to be 0.1068 s using the Dynamic Range Quantized TFLite version of the baseline model, with slight degradation in accuracy. For the segmentation app, the times are 1.8241, 5.2641, 6.2162, and 3.2023 s, respectively, when using JPEG inputs. The Jetson Xavier NX and Android phone stand out as the best platforms due to their compact size, fast inference times, and affordability.

Neurological disorders are debilitating diseases and cause significant morbidity worldwide, with some resulting in mortality. Magnetic Resonance Imagining (MRI) is the prime modality to evaluate most of the diseases involving the brain. The organic diseases of the brain show lesions which are well-appreciated on MRI, but require radiologists and medical experts for delineation. This is crucial for the differential diagnosis of diseases producing similar plaque patterns on the brain. Artificial Intelligence (AI) techniques can help in automatic brain lesion segmentation using massive publicly available data to train Computer-Aided Differential Diagnosis (CADD) algorithms. The accuracy of such CADD algorithms hugely relies on the accuracy of lesion segmentation Deep Learning (DL) models. In this research, DeepLabV3+ architecture is used for semantic segmentation of brain lesions using multiple publicly available datasets. In order to enhance the accuracy, additional ground truth (GT) lesion masks from MICCAI-21 dataset were obtained from a consultant radiologist, and used for training and testing. In addition, the developed algorithm underwent testing using 5 Multiple Sclerosis (MS) and 5 Neuromyelitis Optica (NMO) cases obtained from UiTM Hospital, and 35 MS and 27 NMO cases from HCTM Malaysia, and annotated by radiologists. The Dice score of the trained DL model on test data from MICCAI-21, MICCAI-16, Baghdad Teaching Hospital dataset, HCTM, and UiTM data is 0.7304, 0.6426, 0.4117, 0.5308, and 0.4951, respectively. The model is embedded in an app called NeuroImaging Lesion Extractor (NILE) and is available for public use. This app can serve as an assistive tool for experts in developing differential diagnosis algorithms for demyelinating diseases like MS and NMO.

Brain extraction is an important preprocessing step in Computer Aided Diagnosis (CAD) from brain MRI. It facilitates stripping off irrelevant extra-cranial tissues including skull, eyes, and neck muscles, thereby enhancing accuracy of inference of an AI CAD system. Numerous studies have been conducted to achieve this task ranging from old-school image processing techniques to more sophisticated deep learning approaches. This research investigates the performance of two deep learning-based semantic segmentation techniques, DeepLabV3+ and U-Net for brain extraction of brain MRI scanning images. The two networks were trained using the largest human brain MRI images dataset having more than 160 thousand training images and the performance is compared with SynthStrip from MIT, as the current state of the art system. Besides, the trained networks were also tested using newly collected MRI images from a private hospital, AIH, Islamabad. The result indicates that DeepLabV3+ outperforms SynthStrip and U-Net in all public datasets and produces comparable results from the new dataset with a mean Dice score of 0.98. Using the trained DeepLabV3+, an app called NIVE is developed for public use. Since the DeepLabV3+ was trained using the most comprehensive human brain MRI dataset to date, NIVE is essentially the most versatile brain extraction app capable of handling MRI images in all types of file formats, sequences, orientations, acquisition hardware, and subject age.