Background selection schema on deep learning-based classification of dermatological disease

Multi-task learning for segmentation and classification of breast tumors from ultrasound images

Segmentation and classification of breast tumors are critical components of breast ultrasound (BUS) computer-aided diagnosis (CAD), which significantly improves the diagnostic accuracy of breast cancer. However, the characteristics of tumor regions in BUS images, such as non-uniform intensity distributions, ambiguous or missing boundaries, and varying tumor shapes and sizes, pose significant challenges to automated segmentation and classification solutions. Many previous studies have proposed multi-task learning methods to jointly tackle tumor segmentation and classification by sharing the features extracted by the encoder. Unfortunately, this often introduces redundant or misleading information, which hinders effective feature exploitation and adversely affects performance. To address this issue, we present ACSNet, a novel multi-task learning network designed to optimize tumor segmentation and classification in BUS images. The segmentation network incorporates a novel gate unit to allow optimal transfer of valuable contextual information from the encoder to the decoder. In addition, we develop the Deformable Spatial Attention Module (DSAModule) to improve segmentation accuracy by overcoming the limitations of conventional convolution in dealing with morphological variations of tumors. In the classification branch, multi-scale feature extraction and channel attention mechanisms are integrated to discriminate between benign and malignant breast tumors. Experiments on two publicly available BUS datasets demonstrate that ACSNet not only outperforms mainstream multi-task learning methods for both breast tumor segmentation and classification tasks, but also achieves state-of-the-art results for BUS tumor segmentation. Code and models are available at https://github.com/qqhe-frank/BUS-segmentation-and-classification.git.

Vocal Cord Leukoplakia Classification Using Siamese Network Under Small Samples of White Light Endoscopy Images

OBJECTIVE

Accurate vocal cord leukoplakia classification is instructive for clinical diagnosis and surgical treatment. This article introduces a reliable very deep Siamese network for accurate vocal cord leukoplakia classification.

STUDY DESIGN

A study of a classification network based on a retrospective database.

SETTING

Academic university and hospital.

METHODS

The white light image datasets of vocal cord leukoplakia used in this article were classified into 6 classes: normal tissues, inflammatory keratosis, mild dysplasia, moderate dysplasia, severe dysplasia, and squamous cell carcinoma. The classification performance was assessed by comparing it with 6 classical deep learning models, including AlexNet, VGG Net, Google Inception, ResNet, DenseNet, and Vision Transformer.

RESULTS

Experiments show the superior classification performance of our proposed network compared to state-of-the-art methods. The overall accuracy is 0.9756. The values of sensitivity and specificity are very high as well. The confusion matrix provides information for the 6-class classification task and demonstrates the superiority of our proposed network.

CONCLUSION

Our very deep Siamese network can provide accurate classification results of vocal cord leukoplakia, which facilitates early detection, clinical diagnosis, and surgical treatment. The excellent performance obtained in white light images can reduce the cost for patients, especially those living in developing countries.

Pseudo anomalies enhanced deep support vector data description for electrocardiogram quality assessment

Electrocardiogram (ECG) recordings obtained from wearable devices are susceptible to noise interference that degrades the signal quality. Traditional methods for assessing the quality of electrocardiogram signals (SQA) are mostly supervised and typically rely on limited types of noise in the training data, which imposes limitations in detecting unknown anomalies. The high variability of both ECG signals and noise presents a greater challenge to the generalization of traditional methods. In this paper, we propose a simple and effective unsupervised SQA method by modeling the SQA of ECG as a problem of anomaly detection, in which, a model of pseudo anomalies enhanced deep support vector data description is introduced to learn a more discriminative and generalized hypersphere of the high-quality ECG in a self-supervised manner. Specifically, we propose a series of ECG noise-generation methods to simulate the noise of real scenarios and use the generated noise samples as the pseudo anomalies to correct the hypersphere learned solely by the high-quality ECG samples. Finally, the quality of ECG can be measured based on the distance to the center of the hypersphere. Extensive experimental results on multiple public datasets and our constructed real-world 12-lead dataset demonstrate the effectiveness of the proposed method.

Bridging the gap: Geometry-centric discriminative manifold distribution alignment for enhanced classification in colorectal cancer imaging

The early detection of colorectal cancer (CRC) through medical image analysis is a pivotal concern in healthcare, with the potential to significantly reduce mortality rates. Current Domain Adaptation (DA) methods strive to mitigate the discrepancies between different imaging modalities that are critical in identifying CRC, yet they often fall short in addressing the complexity of cancer's presentation within these images. These conventional techniques typically overlook the intricate geometrical structures and the local variations within the data, leading to suboptimal diagnostic performance. This study introduces an innovative application of the Discriminative Manifold Distribution Alignment (DMDA) method, which is specifically engineered to enhance the medical image diagnosis of colorectal cancer. DMDA transcends traditional DA approaches by focusing on both local and global distribution alignments and by intricately learning the intrinsic geometrical characteristics present in manifold space. This is achieved without depending on the potentially misleading pseudo-labels, a common pitfall in existing methodologies. Our implementation of DMDA on three distinct datasets, involving several unique DA tasks, has consistently demonstrated superior classification accuracy and computational efficiency. The method adeptly captures the complex morphological and textural nuances of CRC lesions, leading to a significant leap in domain adaptation technology. DMDA's ability to reconcile global and local distributional disparities, coupled with its manifold-based geometrical structure learning, signals a paradigm shift in medical imaging analysis. The results obtained are not only promising in terms of advancing domain adaptation theory but also in their practical implications, offering the prospect of substantially improved diagnostic accuracy and faster clinical workflows. This heralds a transformative approach in personalized oncology care, aligning with the pressing need for early and accurate CRC detection.

Medical image identification methods: A review

The identification of medical images is an essential task in computer-aided diagnosis, medical image retrieval and mining. Medical image data mainly include electronic health record data and gene information data, etc. Although intelligent imaging provided a good scheme for medical image analysis over traditional methods that rely on the handcrafted features, it remains challenging due to the diversity of imaging modalities and clinical pathologies. Many medical image identification methods provide a good scheme for medical image analysis. The concepts pertinent of methods, such as the machine learning, deep learning, convolutional neural networks, transfer learning, and other image processing technologies for medical image are analyzed and summarized in this paper. We reviewed these recent studies to provide a comprehensive overview of applying these methods in various medical image analysis tasks, such as object detection, image classification, image registration, segmentation, and other tasks. Especially, we emphasized the latest progress and contributions of different methods in medical image analysis, which are summarized base on different application scenarios, including classification, segmentation, detection, and image registration. In addition, the applications of different methods are summarized in different application area, such as pulmonary, brain, digital pathology, brain, skin, lung, renal, breast, neuromyelitis, vertebrae, and musculoskeletal, etc. Critical discussion of open challenges and directions for future research are finally summarized. Especially, excellent algorithms in computer vision, natural language processing, and unmanned driving will be applied to medical image recognition in the future.

Digital healthcare framework for patients with disabilities based on deep federated learning schemes

Utilizing digital healthcare services for patients who use wheelchairs is a vital and effective means to enhance their healthcare. Digital healthcare integrates various healthcare facilities, including local laboratories and centralized hospitals, to provide healthcare services for individuals in wheelchairs. In digital healthcare, the Internet of Medical Things (IoMT) allows local wheelchairs to connect with remote digital healthcare services and generate sensors from wheelchairs to monitor and process healthcare. Recently, it has been observed that wheelchair patients, when older than thirty, suffer from high blood pressure, heart disease, body glucose, and others due to less activity because of their disabilities. However, existing wheelchair IoMT applications are straightforward and do not consider the healthcare of wheelchair patients with their diseases during their disabilities. This paper presents a novel digital healthcare framework for patients with disabilities based on deep-federated learning schemes. In the proposed framework, we offer the federated learning deep convolutional neural network schemes (FL-DCNNS) that consist of different sub-schemes. The offloading scheme collects the sensors from integrated wheelchair bio-sensors as smartwatches such as blood pressure, heartbeat, body glucose, and oxygen. The smartwatches worked with wearable devices for disabled patients in our framework. We present the federated learning-enabled laboratories for data training and share the updated weights with the data security to the centralized node for decision and prediction. We present the decision forest for centralized healthcare nodes to decide on aggregation with the different constraints: cost, energy, time, and accuracy. We implemented a deep CNN scheme in each laboratory to train and validate the model locally on the node with the consideration of resources. Simulation results show that FL-DCNNS obtained the optimal results on the sensor data and minimized the energy by 25%, time 19%, cost 28%, and improved the accuracy of disease prediction by 99% as compared to existing digital healthcare schemes for wheelchair patients.

Deep learning for medicinal plant species classification and recognition: a systematic review

Knowledge of medicinal plant species is necessary to preserve medicinal plants and safeguard biodiversity. The classification and identification of these plants by botanist experts are complex and time-consuming activities. This systematic review's main objective is to systematically assess the prior research efforts on the applications and usage of deep learning approaches in classifying and recognizing medicinal plant species. Our objective was to pinpoint systematic reviews following the PRISMA guidelines related to the classification and recognition of medicinal plant species through the utilization of deep learning techniques. This review encompassed studies published between January 2018 and December 2022. Initially, we identified 1644 studies through title, keyword, and abstract screening. After applying our eligibility criteria, we selected 31 studies for a thorough and critical review. The main findings of this reviews are (1) the selected studies were carried out in 16 different countries, and India leads in paper contributions with 29%, followed by Indonesia and Sri Lanka. (2) A private dataset has been used in 67.7% of the studies subjected to image augmentation and preprocessing techniques. (3) In 96.7% of the studies, researchers have employed plant leaf organs, with 74% of them utilizing leaf shapes for the classification and recognition of medicinal plant species. (4) Transfer learning with the pre-trained model was used in 83.8% of the studies as a future extraction technique. (5) Convolutional Neural Network (CNN) is used by 64.5% of the paper as a deep learning classifier. (6) The lack of a globally available and public dataset need for medicinal plants indigenous to a specific country and the trustworthiness of the deep learning approach for the classification and recognition of medicinal plants is an observable research gap in this literature review. Therefore, further investigations and collaboration between different stakeholders are required to fulfilling the aforementioned research gaps.

OII-DS: A benchmark Oral Implant Image Dataset for object detection and image classification evaluation

In recent years, there is been a growing reliance on image analysis methods to bolster dentistry practices, such as image classification, segmentation and object detection. However, the availability of related benchmark datasets remains limited. Hence, we spent six years to prepare and test a bench Oral Implant Image Dataset (OII-DS) to support the work in this research domain. OII-DS is a benchmark oral image dataset consisting of 3834 oral CT imaging images and 15240 oral implant images. It serves the purpose of object detection and image classification. To demonstrate the validity of the OII-DS, for each function, the most representative algorithms and metrics are selected for testing and evaluation. For object detection, five object detection algorithms are adopted to test and four evaluation criteria are used to assess the detection of each of the five objects. Additionally, mean average precision serves as the evaluation metric for multi-objective detection. For image classification, 13 classifiers are used for testing and evaluating each of the five categories by meeting four evaluation criteria. Experimental results affirm the high quality of our data in OII-DS, rendering it suitable for evaluating object detection and image classification methods. Furthermore, OII-DS is openly available at the URL for non-commercial purpose: https://doi.org/10.6084/m9.figshare.22608790.

Hierarchical cumulative network for unsupervised medical image registration

Unsupervised deep learning techniques have gained increasing popularity in deformable medical image registration However, existing methods usually overlook the optimal similarity position between moving and fixed images To tackle this issue, we propose a novel hierarchical cumulative network (HCN), which explicitly considers the optimal similarity position with an effective Bidirectional Asymmetric Registration Module (BARM). The BARM simultaneously learns two asymmetric displacement vector fields (DVFs) to optimally warp both moving images and fixed images to their optimal similar shape along the geodesic path. Furthermore, we incorporate the BARM into a Laplacian pyramid network with hierarchical recursion, in which the moving image at the lowest level of the pyramid is warped successively for aligning to the fixed image at the lowest level of the pyramid to capture multiple DVFs. We then accumulate these DVFs and up-sample them to warp the moving images at higher levels of the pyramid to align to the fixed image of the top level. The entire system is end-to-end and jointly trained in an unsupervised manner. Extensive experiments were conducted on two public 3D Brain MRI datasets to demonstrate that our HCN outperforms both the traditional and state-of-the-art registration methods. To further evaluate the performance of our HCN, we tested it on the validation set of the MICCAI Learn2Reg 2021 challenge. Additionally, a cross-dataset evaluation was conducted to assess the generalization of our HCN. Experimental results showed that our HCN is an effective deformable registration method and achieves excellent generalization performance.

Attention2Minority: A salient instance inference-based multiple instance learning for classifying small lesions in whole slide images

Multiple instance learning (MIL) models have achieved remarkable success in analyzing whole slide images (WSIs) for disease classification problems. However, with regard to giga-pixel WSI classification problems, current MIL models are often incapable of differentiating a WSI with extremely small tumor lesions. This minute tumor-to-normal area ratio in a MIL bag inhibits the attention mechanism from properly weighting the areas corresponding to minor tumor lesions. To overcome this challenge, we propose salient instance inference MIL (SiiMIL), a weakly-supervised MIL model for WSI classification. We introduce a novel representation learning for histopathology images to identify representative normal keys. These keys facilitate the selection of salient instances within WSIs, forming bags with high tumor-to-normal ratios. Finally, an attention mechanism is employed for slide-level classification based on formed bags. Our results show that salient instance inference can improve the tumor-to-normal area ratio in the tumor WSIs. As a result, SiiMIL achieves 0.9225 AUC and 0.7551 recall on the Camelyon16 dataset, which outperforms the existing MIL models. In addition, SiiMIL can generate tumor-sensitive attention heatmaps that is more interpretable to pathologists than the widely used attention-based MIL method. Our experiments imply that SiiMIL can accurately identify tumor instances, which could only take up less than 1% of a WSI, so that the ratio of tumor to normal instances within a bag can increase by two to four times.

Machine learning prediction model based on enhanced bat algorithm and support vector machine for slow employment prediction

The employment of college students is an important issue that affects national development and social stability. In recent years, the increase in the number of graduates, the pressure of employment, and the epidemic have made the phenomenon of 'slow employment' increasingly prominent, becoming an urgent problem to be solved. Data mining and machine learning methods are used to analyze and predict the employment prospects for graduates and provide effective employment guidance and services for universities, governments, and graduates. It is a feasible solution to alleviate the problem of 'slow employment' of graduates. Therefore, this study proposed a feature selection prediction model (bGEBA-SVM) based on an improved bat algorithm and support vector machine by extracting 1694 college graduates from 2022 classes in Zhejiang Province. To improve the search efficiency and accuracy of the optimal feature subset, this paper proposed an enhanced bat algorithm based on the Gaussian distribution-based and elimination strategies for optimizing the feature set. The training data were input to the support vector machine for prediction. The proposed method is experimented by comparing it with peers, well-known machine learning models on the IEEE CEC2017 benchmark functions, public datasets, and graduate employment prediction dataset. The experimental results show that bGEBA-SVM can obtain higher prediction Accuracy, which can reach 93.86%. In addition, further education, student leader experience, family situation, career planning, and employment structure are more relevant characteristics that affect employment outcomes. In summary, bGEBA-SVM can be regarded as an employment prediction model with strong performance and high interpretability.

Autism Spectrum Disorder detection framework for children based on federated learning integrated CNN-LSTM

The incidence of Autism Spectrum Disorder (ASD) among children, attributed to genetics and environmental factors, has been increasing daily. ASD is a non-curable neurodevelopmental disorder that affects children's communication, behavior, social interaction, and learning skills. While machine learning has been employed for ASD detection in children, existing ASD frameworks offer limited services to monitor and improve the health of ASD patients. This paper presents a complex and efficient ASD framework with comprehensive services to enhance the results of existing ASD frameworks. Our proposed approach is the Federated Learning-enabled CNN-LSTM (FCNN-LSTM) scheme, designed for ASD detection in children using multimodal datasets. The ASD framework is built in a distributed computing environment where different ASD laboratories are connected to the central hospital. The FCNN-LSTM scheme enables local laboratories to train and validate different datasets, including Ages and Stages Questionnaires (ASQ), Facial Communication and Symbolic Behavior Scales (CSBS) Dataset, Parents Evaluate Developmental Status (PEDS), Modified Checklist for Autism in Toddlers (M-CHAT), and Screening Tool for Autism in Toddlers and Children (STAT) datasets, on different computing laboratories. To ensure the security of patient data, we have implemented a security mechanism based on advanced standard encryption (AES) within the federated learning environment. This mechanism allows all laboratories to offload and download data securely. We integrate all trained datasets into the aggregated nodes and make the final decision for ASD patients based on the decision process tree. Additionally, we have designed various Internet of Things (IoT) applications to improve the efficiency of ASD patients and achieve more optimal learning results. Simulation results demonstrate that our proposed framework achieves an ASD detection accuracy of approximately 99% compared to all existing ASD frameworks.

A despeckling method for ultrasound images utilizing content-aware prior and attention-driven techniques

The despeckling of ultrasound images contributes to the enhancement of image quality and facilitates precise treatment of conditions such as tumor cancers. However, the use of existing methods for eliminating speckle noise can cause the loss of image texture features, impacting clinical judgment. Thus, maintaining clear lesion boundaries while eliminating speckle noise is a challenging task. This paper presents an innovative approach for denoising ultrasound images using a novel noise reduction network model called content-aware prior and attention-driven (CAPAD). The model employs a neural network to automatically capture the hidden prior features in ultrasound images to guide denoising and embeds the denoiser into the optimization module to simultaneously optimize parameters and noise. Moreover, this model incorporates a content-aware attention module and a loss function that preserves the structural characteristics of the image. These additions enhance the network's capacity to capture and retain valuable information. Extensive qualitative evaluation and quantitative analysis performed on a comprehensive dataset provide compelling evidence of the model's superior denoising capabilities. It excels in noise suppression while successfully preserving the underlying structures within the ultrasound images. Compared to other denoising algorithms, it demonstrates an improvement of approximately 5.88% in PSNR and approximately 3.61% in SSIM. Furthermore, using CAPAD as a preprocessing step for breast tumor segmentation in ultrasound images can greatly improve the accuracy of image segmentation. The experimental results indicate that the utilization of CAPAD leads to a notable enhancement of 10.43% in the AUPRC for breast cancer tumor segmentation.

ADDNS: An asymmetric dual deep network with sharing mechanism for medical image fusion of CT and MR-T2

Medical images with different modalities have different semantic characteristics. Medical image fusion aiming to promotion of the visual quality and practical value has become important in medical diagnostics. However, the previous methods do not fully represent semantic and visual features, and the model generalization ability needs to be improved. Furthermore, the brightness-stacking phenomenon is easy to occur during the fusion process. In this paper, we propose an asymmetric dual deep network with sharing mechanism (ADDNS) for medical image fusion. In our asymmetric model-level dual framework, primal Unet part learns to fuse medical images of different modality into a fusion image, while dual Unet part learns to invert the fusion task for multi-modal image reconstruction. This asymmetry of network settings not only enables the ADDNS to fully extract semantic and visual features, but also reduces the model complexity and accelerates the convergence. Furthermore, the sharing mechanism designed according to task relevance also reduces the model complexity and improves the generalization ability of our model. In the end, we use the intermediate supervision method to minimize the difference between fusion image and source images so as to prevent the brightness-stacking problem. Experimental results show that our algorithm achieves better results on both quantitative and qualitative experiments than several state-of-the-art methods.

Evaluating the effect of spinal cord stimulation on patient with disorders of consciousness: A TMS-EEG study

OBJECTIVE

The application of spinal cord stimulation (SCS) in the treatment of disorders of consciousness (DOC) has attracted attention, but its effect on brain activity is still unknown. Transcranial magnetic stimulation combined with EEG (TMS-EEG) can measure cortical activity, which can evaluate the effect of SCS on DOC.

METHODS

We record 20 DOC patients' CRS-R values and TMS-EEG data before and after one-session SCS (Pre-SCS and Post-SCS). 20 DOC patients including 10 patients with unresponsive wakefulness syndrome (UWS) and 10 patients with minimally conscious states (MCS). TMS evoked potential (TEP) was used to measure the changes of cortical activity in DOC patients between Pre-SCS and Post-SCS. Firstly, we used the global mean field potential (GMFP) and fast perturbational complexity index (PCIst) to compare the temporal changes of patients' cortical activity. Then, we obtained the frequency feature (natural frequency, NF) based on the TEP time-frequency analysis, and compared the changes of natural frequency between Pre-SCS and Post-SCS. Finally, the study explored the relationship between the patient's baseline CRS-R values and changes of TMS evoked cortical activity in time and frequency domains.

RESULTS

After SCS, MCS and UWS groups almost have no changes of CRS-R values (MCS: 9.9 ± 1.52 at Pre-SCS, 10.2 ± 1.48 at Post-SCS; UWS: 5.6 ± 1.26 at Pre-SCS, 5.7 ± 1.34 at Post-SCS). MCS group showed significant increases of GMFP amplitude (around 100 ms and 300 ms) and PCIst values at Post-SCS (p < 0.05). UWS group had no significant changes (p > 0.05). Besides, SCS induced the significant increases of natural frequency for MCS group(p < 0.05), but not for UWS group. At last, the study found that all patient's baseline CRS-R values were significantly correlated with ΔPCIst (r = 0.67, p < 0.005), and ΔNF (r = 0.72, p < 0.001).

CONCLUSIONS

SCS can modulate cortical activity of DOC patient, including temporal complexity and natural frequency. The changes of cortical activity caused by SCS are related to patients' consciousness level. TMS-EEG can evaluate the effect of SCS on DOC patients.

Confidence-guided mask learning for semi-supervised medical image segmentation

Semi-supervised learning aims to train a high-performance model with a minority of labeled data and a majority of unlabeled data. Existing methods mostly adopt the mechanism of prediction task to obtain precise segmentation maps with the constraints of consistency or pseudo-labels, whereas the mechanism usually fails to overcome confirmation bias. To address this issue, in this paper, we propose a novel Confidence-Guided Mask Learning (CGML) for semi-supervised medical image segmentation. Specifically, on the basis of the prediction task, we further introduce an auxiliary generation task with mask learning, which intends to reconstruct the masked images for extremely facilitating the model capability of learning feature representations. Moreover, a confidence-guided masking strategy is developed to enhance model discrimination in uncertain regions. Besides, we introduce a triple-consistency loss to enforce a consistent prediction of the masked unlabeled image, original unlabeled image and reconstructed unlabeled image for generating more reliable results. Extensive experiments on two datasets demonstrate that our proposed method achieves remarkable performance.

Self-supervised deep learning for joint 3D low-dose PET/CT image denoising

Deep learning (DL)-based denoising of low-dose positron emission tomography (LDPET) and low-dose computed tomography (LDCT) has been widely explored. However, previous methods have focused only on single modality denoising, neglecting the possibility of simultaneously denoising LDPET and LDCT using only one neural network, i.e., joint LDPET/LDCT denoising. Moreover, DL-based denoising methods generally require plenty of well-aligned LD-normal-dose (LD-ND) sample pairs, which can be difficult to obtain. To this end, we propose a self-supervised two-stage training framework named MAsk-then-Cycle (MAC), to achieve self-supervised joint LDPET/LDCT denoising. The first stage of MAC is masked autoencoder (MAE)-based pre-training and the second stage is self-supervised denoising training. Specifically, we propose a self-supervised denoising strategy named cycle self-recombination (CSR), which enables denoising without well-aligned sample pairs. Unlike other methods that treat noise as a homogeneous whole, CSR disentangles noise into signal-dependent and independent noises. This is more in line with the actual imaging process and allows for flexible recombination of noises and signals to generate new samples. These new samples contain implicit constraints that can improve the network's denoising ability. Based on these constraints, we design multiple loss functions to enable self-supervised training. Then we design a CSR-based denoising network to achieve joint 3D LDPET/LDCT denoising. Existing self-supervised methods generally lack pixel-level constraints on networks, which can easily lead to additional artifacts. Before denoising training, we perform MAE-based pre-training to indirectly impose pixel-level constraints on networks. Experiments on an LDPET/LDCT dataset demonstrate its superiority over existing methods. Our method is the first self-supervised joint LDPET/LDCT denoising method. It does not require any prior assumptions and is therefore more robust.

Determining the clinical applicability of machine learning models through assessment of reporting across skin phototypes and rarer skin cancer types: A systematic review

Machine learning (ML) models for skin cancer recognition may have variable performance across different skin phototypes and skin cancer types. Overall performance metrics alone are insufficient to detect poor subgroup performance. We aimed (1) to assess whether studies of ML models reported results separately for different skin phototypes and rarer skin cancers, and (2) to graphically represent the skin cancer training datasets used by current ML models. In this systematic review, we searched PubMed, Embase and CENTRAL. We included all studies in medical journals assessing an ML technique for skin cancer diagnosis that used clinical or dermoscopic images from 1 January 2012 to 22 September 2021. No language restrictions were applied. We considered rarer skin cancers to be skin cancers other than pigmented melanoma, basal cell carcinoma and squamous cell carcinoma. We identified 114 studies for inclusion. Rarer skin cancers were included by 8/114 studies (7.0%), and results for a rarer skin cancer were reported separately in 1/114 studies (0.9%). Performance was reported across all skin phototypes in 1/114 studies (0.9%), but performance was uncertain in skin phototypes I and VI from minimal representation of the skin phototypes in the test dataset (9/3756 and 1/3756, respectively). For training datasets, although public datasets were most frequently used, with the most widely used being the International Skin Imaging Collaboration (ISIC) archive (65/114 studies, 57.0%), the largest datasets were private. Our review identified that most ML models did not report performance separately for rarer skin cancers and different skin phototypes. A degree of variability in ML model performance across subgroups is expected, but the current lack of transparency is not justifiable and risks models being used inappropriately in populations in whom accuracy is low.