The lack of annotated medical images limits the performance of deep learning models, which usually need large-scale labelled datasets. Few-shot learning techniques can reduce data scarcity issues and enhance medical image analysis speed and robustness. This systematic review gives a comprehensive overview of few-shot learning methods for medical image analysis, aiming to establish a standard methodological pipeline for future research reference. With a particular emphasis on the role of meta-learning, we analysed 80 relevant articles published from 2018 to 2023, conducting a risk of bias assessment and extracting relevant information, especially regarding the employed learning techniques. From this, we delineated a comprehensive methodological pipeline shared among all studies. In addition, we performed a statistical analysis of the studies' results concerning the clinical task and the meta-learning method employed while also presenting supplemental information such as imaging modalities and model robustness evaluation techniques. We discussed the findings of our analysis, providing a deep insight into the limitations of the state-of-the-art methods and the most promising approaches. Drawing on our investigation, we yielded recommendations on potential future research directions aiming to bridge the gap between research and clinical practice.

Vision transformers represent the cutting-edge topic in computer vision and are usually employed on two-dimensional data following a transfer learning approach. In this work, we propose a trained-from-scratch stacking ensemble of 3D-vision transformers to assess prostate cancer aggressiveness from T2-weighted images to help radiologists diagnose this disease without performing a biopsy. We trained 18 3D-vision transformers on T2-weighted axial acquisitions and combined them into two- and three-model stacking ensembles. We defined two metrics for measuring model prediction confidence, and we trained all the ensemble combinations according to a five-fold cross-validation, evaluating their accuracy, confidence in predictions, and calibration. In addition, we optimized the 18 base ViTs and compared the best-performing base and ensemble models by re-training them on a 100-sample bootstrapped training set and evaluating each model on the hold-out test set. We compared the two distributions by calculating the median and the 95% confidence interval and performing a Wilcoxon signed-rank test. The best-performing 3D-vision-transformer stacking ensemble provided state-of-the-art results in terms of area under the receiving operating curve (0.89 [0.61–1]) and exceeded the area under the precision–recall curve of the base model of 22% (p < 0.001). However, it resulted to be less confident in classifying the positive class.

The limited availability of specialized image databases (particularly in hospitals, where tools vary between providers) makes it difficult to train deep learning models. This paper presents a few-shot learning methodology that uses a pre-trained ResNet integrated with an encoder as a backbone to encode conditional shape information for the classification of neonatal resuscitation equipment from less than 100 natural images. The model is also strengthened by incorporating a reliability score, which enriches the prediction with an estimation of classification reliability. The model, whose performance is cross-validated, reached a median accuracy performance of over 99% (and a lower limit of 73.4% for the least accurate model/fold) using only 87 meta-training images. During the test phase on complex natural images, performance was slightly degraded due to a sub-optimal segmentation strategy (FastSAM) required to maintain the real-time inference phase (median accuracy 87.25%). This methodology proves to be excellent for applying complex classification models to contexts (such as neonatal resuscitation) that are not available in public databases. Improvements to the automatic segmentation strategy prior to the extraction of conditional information will allow a natural application in simulation and hospital settings.

Background and objective: Employing deep learning models in critical domains such as medical imaging poses challenges associated with the limited availability of training data. We present a strategy for improving the performance and generalization capabilities of models trained in low-data regimes. Methods: The proposed method starts with a pre-training phase, where features learned in a self-supervised learning setting are disentangled to improve the robustness of the representations for downstream tasks. We then introduce a meta-fine-tuning step, leveraging related classes between meta-training and meta-testing phases but varying the granularity level. This approach aims to enhance the model's generalization capabilities by exposing it to more challenging classification tasks during meta-training and evaluating it on easier tasks but holding greater clinical relevance during meta-testing. We demonstrate the effectiveness of the proposed approach through a series of experiments exploring several backbones, as well as diverse pre-training and fine-tuning schemes, on two distinct medical tasks, i.e., classification of prostate cancer aggressiveness from MRI data and classification of breast cancer malignity from microscopic images. Results: Our results indicate that the proposed approach consistently yields superior performance w.r.t. ablation experiments, maintaining competitiveness even when a distribution shift between training and evaluation data occurs. Conclusion: Extensive experiments demonstrate the effectiveness and wide applicability of the proposed approach. We hope that this work will add another solution to the arsenal of addressing learning issues in data-scarce imaging domains.