Farhad Hoseyni

Abstract:

Intracranial hemorrhages (ICHs) pose a critical medical challenge with a high mortality rate, necessitating timely and accurate diagnosis. This study focuses on enhancing the performance of deep learning models for classifying brain hemorrhages in CT scan images using the PhysioNet dataset. A grid search methodology was employed to tune hyperparameters, particularly addressing the imbalance between ICH and healthy slices. The study utilized ResNet50 for hyperparameter tuning, achieving significant improvements in sensitivity and overall performance through undersampling, bootstrapping, augmentation, and weighted loss techniques. The ResNet50 model demonstrated remarkable performance at the slice-level scope, achieving a sensitivity of 0.94, a specificity of 0.91, an F1 score of 0.64, and an accuracy of 0.91. It outperformed the literature in both sensitivity and accuracy, despite utilizing a smaller and more constrained dataset. At the patient-level scope, our approach achieved even higher metrics with a sensitivity of 1.00, a specificity of 0.80, an F1 score of 0.86, and an accuracy of 0.88, surpassing the literature across all reported metrics.

Abstract:

Data augmentation is a key technique for addressing the challenge of limited datasets, which have become a major component in the training procedures of image processing. Techniques such as geometric transformations and color space adjustments have been thoroughly tested for their ability to artificially expand training datasets and generate semi-realistic data for training purposes. Data augmentation is the most important key to addressing the challenge of limited datasets, which have become a major component of image processing training procedures. Data augmentation techniques, such as geometric transformations and color space adjustments, are thoroughly tested for their ability to artificially expand training datasets and generate semi-realistic data for training purposes. Polygons play a crucial role in instance segmentation and have seen a surge in use across advanced models, such as YOLOv8. Despite their growing popularity, the lack of specialized libraries hampers the polygon-augmentation process. This paper introduces a novel solution to this challenge, embodied in the newly developed AugmenTory library. Notably, AugmenTory offers reduced computational demands in both time and space compared to existing methods. Additionally, the library includes a postprocessing thresholding feature.

Abstract:

In recent years, much research has been focused on dental problem classification, detection, and segmentation using Deep Convolutional Neural Networks (CNN or DCNN) models. Early detection of oral issues is crucial to maintain oral health and prevent the need for invasive treatments. Dental radiographs assist in this by providing detailed images. Integrating intelligent methods with conventional imaging techniques can result in more accurate diagnoses and treatment plans for patients, dental professionals, and service providers. Our study presents a novel unified method that utilizes CNNs to segment dental restorations and cavities in X-ray images. Various datasets, including publicly available one annotated in this study, were used to evaluate the method’s performance. The first step in our method is to detect teeth, identifying each tooth’s number and location in accordance with the Universal Teeth Numbering (UTN) system. This study has also presented a new post-processing technique to address teeth ordering predicting issues in X-ray images. The detection model achieved high precision and recall metrics of over 90% after post-processing, outperforming existing methods. The second step aims to crop images of each tooth from the initial network and precisely segment the cavity and restoration. The segmentation model achieved high precision and recall of over 60%. We proportionately projected the predicted outcomes onto the original image for the final evaluation. The overall performance of our proposed method exceeded that of state-of the-art models. This method detects abnormalities that may go unnoticed by humans, reducing workload and minimizing bias in treatment assessment.

This study is currently in the research and development phase. A collaboration has been established between the APAC research group at KNTU and Iran Medical University to gather a comprehensive dataset for our project. The aim of this project is to detect, classify, and segment intracranial hemorrhages (ICH) in brain CT scans using advanced deep learning techniques.

This study is currently in the research and development phase. A collaboration has been established between the APAC research group at KNTU and Iran Medical University to gather a comprehensive dataset for our project. The aim of this project is to detect, classify, and segment intracranial hemorrhages (ICH) in brain CT scans using advanced deep learning techniques.