Identifying risk factors of intracerebral hemorrhage stability using explainable attention model.

Segmentation of intracerebral hemorrhage (ICH) helps improve the quality of diagnosis, draft the desired treatment methods, and clinically observe the variations with healthy patients. The clinical utilization of various ICH progression scoring systems has limitations due to the systems' modest predictive value. This paper proposes a single pipeline of a multi-task model for end-to-end hemorrhage segmentation and risk estimation. We introduce a 3D spatial attention unit and integrate it into the state-of-the-art segmentation architecture, UNet, to enhance the accuracy by bootstrapping the global spatial representation. We further extract the geometric features from the segmented hemorrhage volume and fuse them with clinical features such as CT angiography (CTA) spot, Glasgow Coma Scale (GCS), and age to predict the ICH stability. Several state-of-the-art machine learning techniques such as multilayer perceptron (MLP), support vector machine (SVM), gradient boosting, and random forests are applied to train stability estimation and to compare the performances. To align clinical intuition with model learning, we determine the shapely values (SHAP) and explain the most significant features for the ICH risk scoring system. A total of 79 patients are included, of which 20 are found in critical condition. Our proposed single pipeline model achieves a segmentation accuracy of 86.3%, stability prediction accuracy of 78.3%, and precision of 82.9%; the mean square error of exact expansion rate regression is observed to be 0.46. The SHAP analysis reveals that CTA spot sign, age, solidity, location, and length of the first axis of the ICH volume are the most critical characteristics that help define the stability of the stroke lesion. We also show that integrating significant geometric features with clinical features can improve the ICH progression scoring by predicting long-term outcomes. Graphical abstract Overview of our proposed method comprising of spatial attention and feature extraction mechanisms. The architecture is trained on the input CT images, and the first step output is the predicted segmentation of the hemorrhagic region. The output is fed into a geometric feature extractor and is fused with clinical features to estimate ICH stability using a multilayer perceptron (MLP).

ST-MTL: Spatio-Temporal multitask learning model to predict scanpath while tracking instruments in robotic surgery.

Representation learning of the task-oriented attention while tracking instrument holds vast potential in image-guided robotic surgery. Incorporating cognitive ability to automate the camera control enables the surgeon to concentrate more on dealing with surgical instruments. The objective is to reduce the operation time and facilitate the surgery for both surgeons and patients. We propose an end-to-end trainable Spatio-Temporal Multi-Task Learning (ST-MTL) model with a shared encoder and spatio-temporal decoders for the real-time surgical instrument segmentation and task-oriented saliency detection. In the MTL model of shared-parameters, optimizing multiple loss functions into a convergence point is still an open challenge. We tackle the problem with a novel asynchronous spatio-temporal optimization (ASTO) technique by calculating independent gradients for each decoder. We also design a competitive squeeze and excitation unit by casting a skip connection that retains weak features, excites strong features, and performs dynamic spatial and channel-wise feature recalibration. To capture better long term spatio-temporal dependencies, we enhance the long-short term memory (LSTM) module by concatenating high-level encoder features of consecutive frames. We also introduce Sinkhorn regularized loss to enhance task-oriented saliency detection by preserving computational efficiency. We generate the task-aware saliency maps and scanpath of the instruments on the dataset of the MICCAI 2017 robotic instrument segmentation challenge. Compared to the state-of-the-art segmentation and saliency methods, our model outperforms most of the evaluation metrics and produces an outstanding performance in the challenge.