AI in cancer detection

This paper develops a Gaussian process-based method for TMB prediction from whole slide images, outperforming baseline methods by incorporating heteroscedastic noise modeling.

This study develops a deep learning method to automatically extract explainable features from histopathological images, achieving higher accuracy than human pathologists in predicting prostate cancer recurrence.

The paper introduces a deep recurrent attention model for histopathological image analysis, showing comparable performance to CNNs while processing fewer patches.

The paper applies transfer learning using ResNet50 to extract diagnostic features from digital pathology images of breast lesions at various magnifications for classification into histological labels.

DeepEM improves pulmonary nodule detection in CT images by using weakly supervised labels from electronic medical records.

This paper presents an automatic method for Gleason score classification of prostate cancer using CSLBP descriptors and SVM, achieving over 79% accuracy on TCGA data.

Fine-tuned ImageNet ResNets beat Inception models on tissue micro-array and BreaKHis slides, reaching 99.8% across four tumour tissues and 96.4% across four malignant sub-types.

The paper quantifies spatial neighborhood preferences of cancer cells in Hodgkin lymphoma using whole slide images and statistical analysis.

The paper identifies tissue morphology biomarkers in long-term PDAC survivors using digital pathology and machine learning, finding reduced heterogeneity and extra-tumoral space features correlate with better survival.

On LIDC-IDRI, gradient boosting on 3D dual path network features reaches 90.44% accuracy on annotated nodules; with a 3D Faster R-CNN detector, patient-level accuracy is 81.41%, 99% of doctors' average.

This paper proposes deep multi-instance networks for whole mammogram classification without requiring manual annotations, showing robustness on the INbreast dataset.

An end-to-end fully convolutional network with a positional prior, conditional random field structural learning, and adversarial training segments mammographic masses on INbreast and DDSM-BCRP.