Understanding skin color bias in deep learning-based skin lesion segmentation

BACKGROUND

The field of dermatological image analysis using deep neural networks includes the semantic segmentation of skin lesions, pivotal for lesion analysis, pathology inference, and diagnoses. While biases in neural network-based dermatoscopic image classification against darker skin tones due to dataset imbalance and contrast disparities are acknowledged, a comprehensive exploration of skin color bias in lesion segmentation models is lacking. It is imperative to address and understand the biases in these models.

METHODS

Our study comprehensively evaluates skin tone bias within prevalent neural networks for skin lesion segmentation. Since no information about skin color exists in widely used datasets, to quantify the bias we use three distinct skin color estimation methods: Fitzpatrick skin type estimation, Individual Typology Angle estimation as well as manual grouping of images by skin color. We assess bias across common models by training a variety of U-Net-based models on three widely-used datasets with 1758 different dermoscopic and clinical images. We also evaluate commonly suggested methods to mitigate bias.

RESULTS

Our findings expose a significant and large correlation between segmentation performance and skin color, revealing consistent challenges in segmenting lesions for darker skin tones across diverse datasets. Using various methods of skin color quantification, we have found significant bias in skin lesion segmentation against darker-skinned individuals when evaluated both in and out-of-sample. We also find that commonly used methods for bias mitigation do not result in any significant reduction in bias.

CONCLUSIONS

Our findings suggest a pervasive bias in most published lesion segmentation methods, given our use of commonly employed neural network architectures and publicly available datasets. In light of our findings, we propose recommendations for unbiased dataset collection, labeling, and model development. This presents the first comprehensive evaluation of fairness in skin lesion segmentation.

Artificial intelligence in forensic medicine and forensic dentistry

This review article aims to highlight the current possibilities for applying Artificial Intelligence in modern forensic medicine and forensic dentistry and present the advantages and disadvantages of its use. For this purpose, the relevant academic literature was searched using PubMed, Web of Science and Scopus. The application of Artificial Intelligence in forensic medicine and forensic dentistry is still in its early stages. However, the possibilities are great, and the future will show what is applicable in daily practice. Artificial Intelligence will improve the accuracy and efficiency of work in forensic medicine and forensic dentistry; it can automate some tasks; and enhance the quality of evidence. Disadvantages of the application of Artificial Intelligence may be related to discrimination, transparency, accountability, privacy, security, ethics and others. Artificial Intelligence systems should be used as a support tool, not as a replacement for forensic experts.

Segment-then-Segment: Context-Preserving Crop-Based Segmentation for Large Biomedical Images

Medical images are often of huge size, which presents a challenge in terms of memory requirements when training machine learning models. Commonly, the images are downsampled to overcome this challenge, but this leads to a loss of information. We present a general approach for training semantic segmentation neural networks on much smaller input sizes called Segment-then-Segment. To reduce the input size, we use image crops instead of downscaling. One neural network performs the initial segmentation on a downscaled image. This segmentation is then used to take the most salient crops of the full-resolution image with the surrounding context. Each crop is segmented using a second specially trained neural network. The segmentation masks of each crop are joined to form the final output image. We evaluate our approach on multiple medical image modalities (microscopy, colonoscopy, and CT) and show that this approach greatly improves segmentation performance with small network input sizes when compared to baseline models trained on downscaled images, especially in terms of pixel-wise recall.

Comparison of the third molar maturity index (I3M) between left and right lower third molars to assess the age of majority: a multi-ethnic study sample

The diagnostic accuracy of the I_3M to assess the legal age of 18 years has already been tested in several specific-population samples. The left lower third molar has been extensively used for discriminating between minors and adults. This research aimed to compare the usefulness of lower third molar maturity indexes, from both left and right side (I_3ML and I_3MR), in samples originating from four distinct continents in order to examine possible differences in their accuracy values. For this purpose, a sample of 10,181 orthopantomograms (OPGs), from Europe, Africa, Asia and America, was analysed and previously scored in other studies. The samples included healthy subjects with no systemic disorders with both third molars and clear depicted root apices. Wilcoxon Signed Rank test for left and right asymmetry did not show any significant differences. Data about sensitivity, specificity, predictive values, likelihood ratio and accuracy were pooled together and showed similar results for I_3ML and I_3MR, respectively. In addition, all these quantities were high when only the I_3MR was considered to discriminate between adults and minors. The present referable database was the first to pool third molar measurements using panoramic radiographs of subjects coming from different continents. The results highlighted that both I_3ML and I_3MR are reliable indicators for assessing the legal age of 18 years old in those jurisdictions where this legal threshold has been set as the age of majority.

Carabelli's trait in Croatian populations over 1800 years

Examination and comparison of the morphological features of tooth crown in archaeological and recent samples can be difficult due to the different levels of tooth wear seen both within and between populations. These differences make the comparison of frequency data for Carabelli trait problematic. The aim of the present study is to detect the frequency and degree of expression of Carabelli's trait in Croatian populations from late antiquity to recent times and to use these data as supplementary evidence of complex population migration. A total of 1287 individuals from the late antiquity, medieval, early modern and modern periods were examined. Correlation between the presence of Carabelli's trait and tooth crown size was tested. The results of our analyses show that the frequency of Carabelli's trait is significantly greater in the early modern period (51.3%) and in the 21st century (43.1%) than in the late antiquity (20.4%) and medieval periods (23.4%). These results are consistent with historical evidence of migration and population change in the territory of present-day Croatia throughout the almost 1800 years covered by this study. The results also provide additional evidence for the complex nature of population change in the transition from the late antiquity to the early medieval period.

Age estimation in archaeological skeletal remains: evaluation of four non-destructive age calculation methods

Estimation of age at death is an essential part of reconstructing information from skeletal material. The aim of the investigation was to reconstruct the chronological age of an archaeological sample from Croatia using cranial skeletal remains as well as to make an evaluation of the methods used for age estimation. For this purpose, four age calculation methods were used: palatal suture closure, occlusal tooth wear, tooth root translucency and pulp/tooth area ratio. Cramer's V test was used to test the association between the age calculation methods. Cramer's V test showed high association (0.677) between age determination results using palatal suture closure and occlusal tooth wear, and low association (0.177) between age determination results using palatal suture closure and pulp/tooth area ratio. Simple methods like palatal suture closure can provide data about age at death for large number of individuals, but with less accuracy. More complex methods which require qualified and trained personnel can provide data about age for a smaller number of individuals, but with more accuracy. Using different (both simple and complex) age calculation methods in archaeological samples can raise the level of confidence and percentage of success in determining age.