Automated CT LI-RADS v2018 scoring of liver observations using machine learning: A multivendor, multicentre retrospective study.

Background & Aims: Assessment of computed tomography (CT)/magnetic resonance imaging Liver Imaging Reporting and Data System (LI-RADS) v2018 major features leads to substantial inter-reader variability and potential decrease in hepatocellular carcinoma diagnostic accuracy. We assessed the performance and added-value of a machine learning (ML)-based algorithm in assessing CT LI-RADS major features and categorisation of liver observations compared with qualitative assessment performed by a panel of radiologists. Methods: High-risk patients as per LI-RADS v2018 with pathologically proven liver lesions who underwent multiphase contrast-enhanced CT at diagnosis between January 2015 and March 2019 in seven centres in five countries were retrospectively included and randomly divided into a training set (n = 84 lesions) and a test set (n = 345 lesions). An ML algorithm was trained to classify non-rim arterial phase hyperenhancement, washout, and enhancing capsule as present, absent, or of uncertain presence. LI-RADS major features and categories were compared with qualitative assessment of two independent readers. The performance of a sequential use of the ML algorithm and independent readers were also evaluated in a triage and an add-on scenario in LR-3/4 lesions. The combined evaluation of three other senior readers was used as reference standard. Results: A total of 318 patients bearing 429 lesions were included. Sensitivity and specificity for LR-5 in the test set were 0.67 (95% CI, 0.62-0.72) and 0.91 (95% CI, 0.87-0.96) respectively, with 242 (70.1%) lesions accurately categorised. Using the ML algorithm in a triage scenario improved the overall performance for LR-5. (0.86 and 0.93 sensitivity, 0.82 and 0.76 specificity, 78% and 82.3% accuracy for the two independent readers). Conclusions: Quantitative assessment of CT LI-RADS v2018 major features is feasible and diagnoses LR-5 observations with high performance especially in combination with the radiologist's visual analysis in patients at high-risk for HCC. Impact and implications: Assessment of CT/MRI LI-RADS v2018 major features leads to substantial inter-reader variability and potential decrease in hepatocellular carcinoma diagnostic accuracy. Rather than replacing radiologists, our results highlight the potential benefit from the radiologist-artificial intelligence interaction in improving focal liver lesions characterisation by using the developed algorithm as a triage tool to the radiologist's visual analysis. Such an AI-enriched diagnostic pathway may help standardise and improve the quality of analysis of liver lesions in patients at high risk for HCC, especially in non-expert centres in liver imaging. It may also impact the clinical decision-making and guide the clinician in identifying the lesions to be biopsied, for instance in patients with multiple liver focal lesions.

Multi-task multi-scale learning for outcome prediction in 3D PET images.

BACKGROUND AND OBJECTIVES: Predicting patient response to treatment and survival in oncology is a prominent way towards precision medicine. To this end, radiomics has been proposed as a field of study where images are used instead of invasive methods. The first step in radiomic analysis in oncology is lesion segmentation. However, this task is time consuming and can be physician subjective. Automated tools based on supervised deep learning have made great progress in helping physicians. However, they are data hungry, and annotated data remains a major issue in the medical field where only a small subset of annotated images are available. METHODS: In this work, we propose a multi-task, multi-scale learning framework to predict patient's survival and response. We show that the encoder can leverage multiple tasks to extract meaningful and powerful features that improve radiomic performance. We also show that subsidiary tasks serve as an inductive bias so that the model can better generalize. RESULTS: Our model was tested and validated for treatment response and survival in esophageal and lung cancers, with an area under the ROC curve of 77% and 71% respectively, outperforming single-task learning methods. CONCLUSIONS: Multi-task multi-scale learning enables higher performance of radiomic analysis by extracting rich information from intratumoral and peritumoral regions.

Weakly Supervised Tumor Detection in PET Using Class Response for Treatment Outcome Prediction.

It is proven that radiomic characteristics extracted from the tumor region are predictive. The first step in radiomic analysis is the segmentation of the lesion. However, this task is time consuming and requires a highly trained physician. This process could be automated using computer-aided detection (CAD) tools. Current state-of-the-art methods are trained in a supervised learning setting, which requires a lot of data that are usually not available in the medical imaging field. The challenge is to train one model to segment different types of tumors with only a weak segmentation ground truth. In this work, we propose a prediction framework including a 3D tumor segmentation in positron emission tomography (PET) images, based on a weakly supervised deep learning method, and an outcome prediction based on a 3D-CNN classifier applied to the segmented tumor regions. The key step is to locate the tumor in 3D. We propose to (1) calculate two maximum intensity projection (MIP) images from 3D PET images in two directions, (2) classify the MIP images into different types of cancers, (3) generate the class activation maps through a multitask learning approach with a weak prior knowledge, and (4) segment the 3D tumor region from the two 2D activation maps with a proposed new loss function for the multitask. The proposed approach achieves state-of-the-art prediction results with a small data set and with a weak segmentation ground truth. Our model was tested and validated for treatment response and survival in lung and esophageal cancers on 195 patients, with an area under the receiver operating characteristic curve (AUC) of 67% and 59%, respectively, and a dice coefficient of 73% and 0.77% for tumor segmentation.

Parsimonious path openings and closings.

Path openings and closings are morphological tools used to preserve long, thin, and tortuous structures in gray level images. They explore all paths from a defined class, and filter them with a length criterion. However, most paths are redundant, making the process generally slow. Parsimonious path openings and closings are introduced in this paper to solve this problem. These operators only consider a subset of the paths considered by classical path openings, thus achieving a substantial speed-up, while obtaining similar results. In addition, a recently introduced 1D opening algorithm is applied along each selected path. Its complexity is linear with respect to the number of pixels, independent of the size of the opening. Furthermore, it is fast for any input data accuracy (integer or floating point) and works in stream. Parsimonious path openings are also extended to incomplete paths, i.e., paths containing gaps. Noise-corrupted paths can thus be processed with the same approach and complexity. These parsimonious operators achieve a several orders of magnitude speed-up. Examples are shown for incomplete path openings, where computing times are brought from minutes to tens of milliseconds, while obtaining similar results.