Deep learning and digital pathology powers prediction of HCC development in steatotic liver disease.

BACKGROUND AND AIMS: Identifying patients with steatotic liver disease who are at a high risk of developing HCC remains challenging. We present a deep learning (DL) model to predict HCC development using hematoxylin and eosin-stained whole-slide images of biopsy-proven steatotic liver disease. APPROACH AND RESULTS: We included 639 patients who did not develop HCC for ≥7 years after biopsy (non-HCC class) and 46 patients who developed HCC <7 years after biopsy (HCC class). Paired cases of the HCC and non-HCC classes matched by biopsy date and institution were used for training, and the remaining nonpaired cases were used for validation. The DL model was trained using deep convolutional neural networks with 28,000 image tiles cropped from whole-slide images of the paired cases, with an accuracy of 81.0% and an AUC of 0.80 for predicting HCC development. Validation using the nonpaired cases also demonstrated a good accuracy of 82.3% and an AUC of 0.84. These results were comparable to the predictive ability of logistic regression model using fibrosis stage. Notably, the DL model also detected the cases of HCC development in patients with mild fibrosis. The saliency maps generated by the DL model highlighted various pathological features associated with HCC development, including nuclear atypia, hepatocytes with a high nuclear-cytoplasmic ratio, immune cell infiltration, fibrosis, and a lack of large fat droplets. CONCLUSIONS: The ability of the DL model to capture subtle pathological features beyond fibrosis suggests its potential for identifying early signs of hepatocarcinogenesis in patients with steatotic liver disease.

Deep convolutional neural network-based algorithm for muscle biopsy diagnosis.

Histopathologic evaluation of muscle biopsy samples is essential for classifying and diagnosing muscle diseases. However, the numbers of experienced specialists and pathologists are limited. Although new technologies such as artificial intelligence are expected to improve medical reach, their use with rare diseases, such as muscle diseases, is challenging because of the limited availability of training datasets. To address this gap, we developed an algorithm based on deep convolutional neural networks (CNNs) and collected 4041 microscopic images of 1400 hematoxylin-and-eosin-stained pathology slides stored in the National Center of Neurology and Psychiatry for training CNNs. Our trained algorithm differentiated idiopathic inflammatory myopathies (mostly treatable) from hereditary muscle diseases (mostly non-treatable) with an area under the curve (AUC) of 0.996 and achieved better sensitivity and specificity than the diagnoses done by nine physicians under limited diseases and conditions. Furthermore, it successfully and accurately classified four subtypes of the idiopathic inflammatory myopathies with an average AUC of 0.958 and classified seven subtypes of hereditary muscle disease with an average AUC of 0.936. We also established a method to validate the similarity between the predictions made by the algorithm and the seven physicians using visualization technology and clarified the validity of the predictions. These results support the reliability of the algorithm and suggest that our algorithm has the potential to be used straightforwardly in a clinical setting.

Deep learning shows the capability of high-level computer-aided diagnosis in malignant lymphoma.

A pathological evaluation is one of the most important methods for the diagnosis of malignant lymphoma. A standardized diagnosis is occasionally difficult to achieve even by experienced hematopathologists. Therefore, established procedures including a computer-aided diagnosis are desired. This study aims to classify histopathological images of malignant lymphomas through deep learning, which is a computer algorithm and type of artificial intelligence (AI) technology. We prepared hematoxylin and eosin (H&E) slides of a lesion area from 388 sections, namely, 259 with diffuse large B-cell lymphoma, 89 with follicular lymphoma, and 40 with reactive lymphoid hyperplasia, and created whole slide images (WSIs) using a whole slide system. WSI was annotated in the lesion area by experienced hematopathologists. Image patches were cropped from the WSI to train and evaluate the classifiers. Image patches at magnifications of ×5, ×20, and ×40 were randomly divided into a test set and a training and evaluation set. The classifier was assessed using the test set through a cross-validation after training. The classifier achieved the highest levels of accuracy of 94.0%, 93.0%, and 92.0% for image patches with magnifications of ×5, ×20, and ×40, respectively, in comparison to diffuse large B-cell lymphoma, follicular lymphoma, and reactive lymphoid hyperplasia. Comparing the diagnostic accuracies between the proposed classifier and seven pathologists, including experienced hematopathologists, using the test set made up of image patches with magnifications of ×5, ×20, and ×40, the best accuracy demonstrated by the classifier was 97.0%, whereas the average accuracy achieved by the pathologists using WSIs was 76.0%, with the highest accuracy reaching 83.3%. In conclusion, the neural classifier can outperform pathologists in a morphological evaluation. These results suggest that the AI system can potentially support the diagnosis of malignant lymphoma.

[Development of Deep Convolutional Neural Network-Based Algorithm for Muscle Biopsy Diagnosis].

To diagnose muscle disease, histopathologic evaluation of muscle biopsy is essential. In addition, since myositis has a well-established treatment, an accurate diagnosis is required. However, distinguishing myositis from other muscle diseases is challenging for pathologists. Thus, artificial intelligence is expected to improve medical productivity. Therefore, we developed an algorithm based on deep convolutional neural networks to make the algorithm for muscle biopsy diagnosis. We used 1,400 hematoxylin-and-eosin-stained pathology slides for training and testing. Our trained algorithm achieved better sensitivity and specificity than the diagnoses made by physicians.