Cross-organ, cross-modality transfer learning: feasibility study for segmentation and classification

Optimizing mammography interpretation education: leveraging deep learning for cohort-specific error detection to enhance radiologist training

Purpose

Accurate interpretation of mammograms presents challenges. Tailoring mammography training to reader profiles holds the promise of an effective strategy to reduce these errors. This proof-of-concept study investigated the feasibility of employing convolutional neural networks (CNNs) with transfer learning to categorize regions associated with false-positive (FP) errors within screening mammograms into categories of "low" or "high" likelihood of being a false-positive detection for radiologists sharing similar geographic characteristics.

Approach

Mammography test sets assessed by two geographically distant cohorts of radiologists (cohorts A and B) were collected. FP patches within these mammograms were segmented and categorized as "difficult" or "easy" based on the number of readers committing FP errors. Patches outside 1.5 times the interquartile range above the upper quartile were labeled as difficult, whereas the remaining patches were labeled as easy. Using transfer learning, a patch-wise CNN model for binary patch classification was developed utilizing ResNet as the feature extractor, with modified fully connected layers for the target task. Model performance was assessed using 10-fold cross-validation.

Results

Compared with other architectures, the transferred ResNet-50 achieved the highest performance, obtaining receiver operating characteristics area under the curve values of 0.933 ( ± 0.012 ) and 0.975 ( ± 0.011 ) on the validation sets for cohorts A and B, respectively.

Conclusions

The findings highlight the feasibility of employing CNN-based transfer learning to predict the difficulty levels of local FP patches in screening mammograms for specific radiologist cohort with similar geographic characteristics.

Transfer learning for anatomical structure segmentation in otorhinolaryngology microsurgery

BACKGROUND

Reducing the annotation burden is an active and meaningful area of artificial intelligence (AI) research.

METHODS

Multiple datasets for the segmentation of two landmarks were constructed based on 41 257 labelled images and 6 different microsurgical scenarios. These datasets were trained using the multi-stage transfer learning (TL) methodology.

RESULTS

The multi-stage TL enhanced segmentation performance over baseline (mIOU 0.6892 vs. 0.8869). Besides, Convolutional Neural Networks (CNNs) achieved a robust performance (mIOU 0.8917 vs. 0.8603) even when the training dataset size was reduced from 90% (30 078 images) to 10% (3342 images). When directly applying the weight from one certain surgical scenario to recognise the same target in images of other scenarios without training, CNNs still obtained an optimal mIOU of 0.6190 ± 0.0789.

CONCLUSIONS

Model performance can be improved with TL in datasets with reduced size and increased complexity. It is feasible for data-based domain adaptation among different microsurgical fields.

Comprehensive Assessment of MRI-based Artificial Intelligence Frameworks Performance in the Detection, Segmentation, and Classification of Prostate Lesions Using Open-Source Databases

Numerous MRI-based artificial intelligence (AI) frameworks have been designed for prostate cancer lesion detection, segmentation, and classification via MRI as a result of intrareader and interreader variability that is inherent to traditional interpretation. Open-source data sets have been released with the intention of providing freely available MRIs for the testing of diverse AI frameworks in automated or semiautomated tasks. Here, an in-depth assessment of the performance of MRI-based AI frameworks for detecting, segmenting, and classifying prostate lesions using open-source databases was performed. Among 17 data sets, 12 were specific to prostate cancer detection/classification, with 52 studies meeting the inclusion criteria.

Deep learning algorithm performance in contouring head and neck organs at risk: a systematic review and single-arm meta-analysis

PURPOSE

The contouring of organs at risk (OARs) in head and neck cancer radiation treatment planning is a crucial, yet repetitive and time-consuming process. Recent studies have applied deep learning (DL) algorithms to automatically contour head and neck OARs. This study aims to conduct a systematic review and meta-analysis to summarize and analyze the performance of DL algorithms in contouring head and neck OARs. The objective is to assess the advantages and limitations of DL algorithms in contour planning of head and neck OARs.

METHODS

This study conducted a literature search of Pubmed, Embase and Cochrane Library databases, to include studies related to DL contouring head and neck OARs, and the dice similarity coefficient (DSC) of four categories of OARs from the results of each study are selected as effect sizes for meta-analysis. Furthermore, this study conducted a subgroup analysis of OARs characterized by image modality and image type.

RESULTS

149 articles were retrieved, and 22 studies were included in the meta-analysis after excluding duplicate literature, primary screening, and re-screening. The combined effect sizes of DSC for brainstem, spinal cord, mandible, left eye, right eye, left optic nerve, right optic nerve, optic chiasm, left parotid, right parotid, left submandibular, and right submandibular are 0.87, 0.83, 0.92, 0.90, 0.90, 0.71, 0.74, 0.62, 0.85, 0.85, 0.82, and 0.82, respectively. For subgroup analysis, the combined effect sizes for segmentation of the brainstem, mandible, left optic nerve, and left parotid gland using CT and MRI images are 0.86/0.92, 0.92/0.90, 0.71/0.73, and 0.84/0.87, respectively. Pooled effect sizes using 2D and 3D images of the brainstem, mandible, left optic nerve, and left parotid gland for contouring are 0.88/0.87, 0.92/0.92, 0.75/0.71 and 0.87/0.85.

CONCLUSIONS

The use of automated contouring technology based on DL algorithms is an essential tool for contouring head and neck OARs, achieving high accuracy, reducing the workload of clinical radiation oncologists, and providing individualized, standardized, and refined treatment plans for implementing "precision radiotherapy". Improving DL performance requires the construction of high-quality data sets and enhancing algorithm optimization and innovation.

A Deep Learning Framework for the Characterization of Thyroid Nodules from Ultrasound Images Using Improved Inception Network and Multi-Level Transfer Learning

In the past few years, deep learning has gained increasingly widespread attention and has been applied to diagnosing benign and malignant thyroid nodules. It is difficult to acquire sufficient medical images, resulting in insufficient data, which hinders the development of an efficient deep-learning model. In this paper, we developed a deep-learning-based characterization framework to differentiate malignant and benign nodules from the thyroid ultrasound images. This approach improves the recognition accuracy of the inception network by combining squeeze and excitation networks with the inception modules. We have also integrated the concept of multi-level transfer learning using breast ultrasound images as a bridge dataset. This transfer learning approach addresses the issues regarding domain differences between natural images and ultrasound images during transfer learning. This paper aimed to investigate how the entire framework could help radiologists improve diagnostic performance and avoid unnecessary fine-needle aspiration. The proposed approach based on multi-level transfer learning and improved inception blocks achieved higher precision (0.9057 for the benign class and 0.9667 for the malignant class), recall (0.9796 for the benign class and 0.8529 for malignant), and F1-score (0.9412 for benign class and 0.9062 for malignant class). It also obtained an AUC value of 0.9537, which is higher than that of the single-level transfer learning method. The experimental results show that this model can achieve satisfactory classification accuracy comparable to experienced radiologists. Using this model, we can save time and effort as well as deliver potential clinical application value.

Modality Bank: Learn multi-modality images across data centers without sharing medical data

Multi-modality images have been widely used and provide comprehensive information for medical image analysis. However, acquiring all modalities among all institutes is costly and often impossible in clinical settings. To leverage more comprehensive multi-modality information, we propose privacy secured decentralized multi-modality adaptive learning architecture named ModalityBank. Our method could learn a set of effective domain-specific modulation parameters plugged into a common domain-agnostic network. We demonstrate by switching different sets of configurations, the generator could output high-quality images for a specific modality. Our method could also complete the missing modalities across all data centers, thus could be used for modality completion purposes. The downstream task trained from the synthesized multi-modality samples could achieve higher performance than learning from one real data center and achieve close- to- real performance compare with all real images.

Transfer learning in medical image segmentation: New insights from analysis of the dynamics of model parameters and learned representations

We present a critical assessment of the role of transfer learning in training fully convolutional networks (FCNs) for medical image segmentation. We first show that although transfer learning reduces the training time on the target task, improvements in segmentation accuracy are highly task/data-dependent. Large improvements are observed only when the segmentation task is more challenging and the target training data is smaller. We shed light on these observations by investigating the impact of transfer learning on the evolution of model parameters and learned representations. We observe that convolutional filters change little during training and still look random at convergence. We further show that quite accurate FCNs can be built by freezing the encoder section of the network at random values and only training the decoder section. At least for medical image segmentation, this finding challenges the common belief that the encoder section needs to learn data/task-specific representations. We examine the evolution of FCN representations to gain a deeper insight into the effects of transfer learning on the training dynamics. Our analysis shows that although FCNs trained via transfer learning learn different representations than FCNs trained with random initialization, the variability among FCNs trained via transfer learning can be as high as that among FCNs trained with random initialization. Moreover, feature reuse is not restricted to the early encoder layers; rather, it can be more significant in deeper layers. These findings offer new insights and suggest alternative ways of training FCNs for medical image segmentation.