Skin cancer is a significant health concern worldwide, and early and accurate diagnosis plays a crucial role in improving patient outcomes. In recent years, deep learning models have shown remarkable success in various computer vision tasks, including image classification. In this research study, we introduce an approach for skin cancer classification using vision transformer, a state-of-the-art deep learning architecture that has demonstrated exceptional performance in diverse image analysis tasks. The study utilizes the HAM10000 dataset; a publicly available dataset comprising 10,015 skin lesion images classified into two categories: benign (6705 images) and malignant (3310 images). This dataset consists of high-resolution images captured using dermatoscopes and carefully annotated by expert dermatologists. Preprocessing techniques, such as normalization and augmentation, are applied to enhance the robustness and generalization of the model. The vision transformer architecture is adapted to the skin cancer classification task. The model leverages the self-attention mechanism to capture intricate spatial dependencies and long-range dependencies within the images, enabling it to effectively learn relevant features for accurate classification. Segment Anything Model (SAM) is employed to segment the cancerous areas from the images; achieving an IOU of 96.01% and Dice coefficient of 98.14% and then various pretrained models are used for classification using vision transformer architecture. Extensive experiments and evaluations are conducted to assess the performance of our approach. The results demonstrate the superiority of the vision transformer model over traditional deep learning architectures in skin cancer classification in general with some exceptions. Upon experimenting on six different models, ViT-Google, ViT-MAE, ViT-ResNet50, ViT-VAN, ViT-BEiT, and ViT-DiT, we found out that the ML approach achieves 96.15% accuracy using Google’s ViT patch-32 model with a low false negative ratio on the test dataset, showcasing its potential as an effective tool for aiding dermatologists in the diagnosis of skin cancer.
Selecting regions of interest (ROI) is a common step in medical image analysis across all imaging modalities. An ROI is a subset of an image appropriate for the intended analysis and identified manually by experts. In modern pathology, the analysis involves processing multidimensional and high resolution whole slide image (WSI) tiles automatically with an overwhelming quantity of structural and functional information. Despite recent improvements in computing capacity, analyzing such a plethora of data is challenging but vital to accurate analysis. Automatic ROI detection can significantly reduce the number of pixels to be processed, speed the analysis, improve accuracy and reduce dependency on pathologists. In this paper, we present an ROI detection method for WSI and demonstrated it for human epidermal growth factor receptor 2 (HER2) grading for breast cancer patients. Existing HER2 grading relies on manual ROI selection, which is tedious, time-consuming and suffers from inter-observer and intra-observer variability. This study found that the HER2 grade changes with ROI selection. We proposed an ROI detection method using Vision Transformer and investigated the role of image magnification for ROI detection. This method yielded an accuracy of 99% using 20 × WSI and 97% using 10 × WSI for the ROI detection. In the demonstration, the proposed method increased the diagnostic agreement to 99.3% with the clinical scores and reduced the time to 15 seconds for automated HER2 grading.
In medical image analysis, segmenting pancreatic CT images presents a significant challenge due to the complex anatomy of the pancreas and the generally low contrast of these images. Accurate pancreas segmentation is crucial in clinical scenarios, particularly for the diagnosis and treatment of pancreatic cancer. The U-Net architecture and its variations have achieved significant progress in deep learning-based image segmentation, especially in the context of pancreatic CT image segmentation. However, there is a noticeable gap in the comprehensive evaluation of their performance, limitations, and potential improvements specifically in this area. This systematic review aims to address this gap in the literature, focusing particularly on U-Net and its variants in pancreatic CT image segmentation. Adhering to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines, this review includes relevant studies published since 2019 in the field of pancreatic segmentation. The findings illuminate the current limitations of these methods and establish a theoretical foundation for future research directions.
Automatic classification of breast cancer in histopathology images is crucial for accurate diagnosis and effective treatment planning. Recently, classification methods based on the ResNet architecture have gained prominence due to their ability to improve accuracy significantly. This is achieved by employing skip connections to mitigate vanishing gradient issues, enabling the integration of low-level and high-level feature information. However, the conventional ResNet architecture faces challenges such as data imbalance and limited interpretability, which necessitate cross-domain knowledge and collaboration among medical experts. To address these challenges, this study proposes a novel method for breast cancer classification: the Dual-Activated Lightweight Attention ResNet50 (DALAResNet50) model. This model integrates a pre-trained ResNet50 architecture with a lightweight attention mechanism, embedding an attention module in the fourth layer of ResNet50, and incorporates two fully connected layers with LeakyReLU and ReLU activation functions to enhance feature learning capabilities. Extensive experiments conducted on the BreakHis, BACH, and Mini-DDSM datasets demonstrate that DALAResNet50 outperforms state-of-the-art models in accuracy, F1 score, IBA, and GMean, particularly excelling in classification tasks involving imbalanced datasets. Furthermore, the proposed Dynamic Threshold Grad-CAM (DT Grad-CAM) method provides clearer and more focused visualizations, enhancing interpretability and assisting medical experts in identifying key features.
Traditionally, pathological analysis and diagnosis are performed by manually eyeballing glass-slide specimens under a microscope by an expert. The whole slide image (WSI) is the digital specimen produced from the glass slide. WSI enabled specimens to be observed on a computer screen and led to computational pathology where computer vision and artificial intelligence are utilized for automated analysis and diagnosis. With the current computational advancement, the entire WSI can be analyzed autonomously without human supervision. However, the analysis could fail or lead to wrong diagnosis if the WSI is affected by tissue artifacts such as tissue fold or air bubbles depending on the severity. Existing artifact detection methods rely on experts for severity assessment to eliminate artifact-affected regions from the analysis. This process is time-consuming, exhausting and undermines the goal of automated analysis or removal of artifacts without evaluating their severity, which could result in the loss of diagnostically important data. Therefore, it is necessary to detect artifacts and then assess their severity automatically. In this paper, we propose a system that incorporates severity evaluation with artifact detection utilizing convolutional neural networks (CNN). The proposed system uses DoubleUNet to segment artifacts and an ensemble network of six fine-tuned CNN models to determine severity. This method outperformed current state-of-the-art in accuracy by 9% for artifact segmentation and achieved a strong correlation of 97% with the pathologist’s evaluation for severity assessment. The robustness of the system was demonstrated using our proposed heterogeneous dataset and practical usability was ensured by integrating it with an automated analysis system.