An Applied Statistics dataset for human vs AI-generated answer classification.

Due to the increasing popularity of Large Language Models (LLMs) like ChatGPT, students from various fields now commonly rely on AI-powered text generation tools to complete their assignments. This poses a challenge for course instructors who struggle to identify the authenticity of submitted work. Several AI detection tools for differentiating human-generated text from AI-generated text exist for domains like medical and coding, and available generic tools do not perform well on domain-specific tasks. Those AI detection tools depend on LLM, and to train the LLM, an instruction dataset is needed that helps the LLM to learn the differences between patterns of human-generated text and AI-generated text. To help with the creation of a tool for Applied Statistics, we have created a dataset containing 4231 question-and-answer combinations. To create the dataset, first, we collected 116 questions covering a wide range of topics from Applied Statistics selected by domain experts. Second, we created a framework to randomly distribute and collect answers to the questions from students. Third, we collected answers to fifty assigned questions from each of the 100 students participating in the work. Fourth, we generated an equal number of AI-generated answers using ChatGPT. The prepared dataset will be useful for creating AI-detector tools for the Applied Statistics domain as well as benchmarking AI-detector tools, and the proposed data preparation framework will be useful for collecting data for other domains.

A skin lesion hair mask dataset with fine-grained annotations.

Occlusion of skin lesions in dermoscopic images due to hair affects the performance of computer-assisted lesion analysis algorithms. Lesion analysis can benefit from digital hair removal or realistic hair simulation techniques. To assist in that process, we have created the largest publicly available skin lesion hair segmentation mask dataset by carefully annotating 500 dermoscopic images. Compared to the existing datasets, our dataset is free of non-hair artifacts like ruler markers, bubbles, and ink marks. The dataset is also less prone to over and under segmentations because of fine-grained annotations and quality checks from multiple independent annotators. To create the dataset, first, we collected five hundred copyright-free CC0 licensed dermoscopic images covering different hair patterns. Second, we trained a deep learning hair segmentation model on a publicly available weakly annotated dataset. Third, we extracted hair masks for the selected five hundred images using the segmentation model. Finally, we manually corrected all the segmentation errors and verified the annotations by superimposing the annotated masks on top of the dermoscopic images. Multiple annotators were involved in the annotation and verification process to make the annotations as error-free as possible. The prepared dataset will be useful for benchmarking and training hair segmentation algorithms as well as creating realistic hair augmentation systems.

Exploring convolutional neural networks with transfer learning for diagnosing Lyme disease from skin lesion images.

BACKGROUND AND OBJECTIVE: Lyme disease which is one of the most common infectious vector-borne diseases manifests itself in most cases with erythema migrans (EM) skin lesions. Recent studies show that convolutional neural networks (CNNs) perform well to identify skin lesions from images. Lightweight CNN based pre-scanner applications for resource-constrained mobile devices can help users with early diagnosis of Lyme disease and prevent the transition to a severe late form thanks to appropriate antibiotic therapy. Also, resource-intensive CNN based robust computer applications can assist non-expert practitioners with an accurate diagnosis. The main objective of this study is to extensively analyze the effectiveness of CNNs for diagnosing Lyme disease from images and to find out the best CNN architectures considering resource constraints. METHODS: First, we created an EM dataset with the help of expert dermatologists from Clermont-Ferrand University Hospital Center of France. Second, we benchmarked this dataset for twenty-three CNN architectures customized from VGG, ResNet, DenseNet, MobileNet, Xception, NASNet, and EfficientNet architectures in terms of predictive performance, computational complexity, and statistical significance. Third, to improve the performance of the CNNs, we used custom transfer learning from ImageNet pre-trained models as well as pre-trained the CNNs with the skin lesion dataset HAM10000. Fourth, for model explainability, we utilized Gradient-weighted Class Activation Mapping to visualize the regions of input that are significant to the CNNs for making predictions. Fifth, we provided guidelines for model selection based on predictive performance and computational complexity. RESULTS: Customized ResNet50 architecture gave the best classification accuracy of 84.42% ±1.36, AUC of 0.9189±0.0115, precision of 83.1%±2.49, sensitivity of 87.93%±1.47, and specificity of 80.65%±3.59. A lightweight model customized from EfficientNetB0 also performed well with an accuracy of 83.13%±1.2, AUC of 0.9094±0.0129, precision of 82.83%±1.75, sensitivity of 85.21% ±3.91, and specificity of 80.89%±2.95. All the trained models are publicly available at https://dappem.limos.fr/download.html, which can be used by others for transfer learning and building pre-scanners for Lyme disease. CONCLUSION: Our study confirmed the effectiveness of even some lightweight CNNs for building Lyme disease pre-scanner mobile applications to assist people with an initial self-assessment and referring them to expert dermatologist for further diagnosis.