傳統(tǒng)濾波方法不保邊的原因是：都使用全窗口回歸，會(huì)有沿著圖像邊緣的擴(kuò)散。本文提出把窗口的邊緣直接放在待處理像素的位置，這就切斷了可能的法線方向的擴(kuò)散。具體到一個(gè)像素位置，直接枚舉八個(gè)可能的方向，讓數(shù)據(jù)自適應(yīng)地選擇一個(gè)最佳的方向。
- 201903 Radiology 人工智能自動(dòng)勾畫鼻咽癌GTV，港中文Pheng-Ann Heng團(tuán)隊(duì) [paper] [deepcare解讀]
Deep Learning for Automated Contouring of Primary Tumor Volumes by MRI for Nasopharyngeal Carcinoma, 818訓(xùn)練，203測(cè)試；用20個(gè)測(cè)試數(shù)據(jù)比較AI和醫(yī)生的分割結(jié)果。AI assistance improved contouring accuracy (five of eight oncologists had a higher median DSC after AI assistance; average median DSC, 0.74 vs 0.78; P < .001), reduced intra- and interobserver variation (by 36.4% and 54.5%, respectively), and reduced contouring time (by 39.4%). AI自動(dòng)勾畫然后醫(yī)生修改，平均精度由74%提高至79%。
201904-騰訊深度解構(gòu)產(chǎn)業(yè)互聯(lián)網(wǎng)：九大領(lǐng)域打法，五個(gè)維度框架，[機(jī)器之心], [騰訊研究院介紹]
201904-前深度學(xué)習(xí)時(shí)代CTR預(yù)估模型的演化之路 [知乎王喆]
- MED NeurIPS 2018: Is your ML Methods solving a real clinical problem? by Tal Arbel
Focus lesion detection, segmentation, disease prediction in patient images
ML in Medical Imaging: patient diagnosis, understanding disease development, predicting patient outcome from images, personalized medicine.
ML方法沒有被廣泛應(yīng)用到臨床workflow的原因/挑戰(zhàn)
- CV中的DL方法在醫(yī)學(xué)圖像中不總是work。比如BraTS分割任務(wù)DL很成功，但是存活時(shí)間預(yù)測(cè)任務(wù)效果不如人意。**Errors in performance lead to clinician mistrust.
- Clinicians don't trust black box methods. Interpretability is very important.
- No large scale annotated medical dataset for training. 導(dǎo)致通常在small, proprietary or benchmark dataset開發(fā)算法，缺乏魯棒性。
Examine machine learning performance and metrics in real clinical contexts
- 臨床影響：將病灶檢測(cè)和分割算法加入商業(yè)軟件中，提升了efficiency and precision，節(jié)省~5倍的時(shí)間和金錢；提升treatment analysis for almost all (22/23) new MS drugs in circulation wordwide. Clinical impact formula: Synergy with clinicians, end-users when designing method + trying methods and metrics for success to real clinical objectives = Clinical impact
201811-MICCAI 18 分割Decathlon冠軍：3D Semi-Supervised Learning with Uncertainty-Aware Multi-View Co-Training，Nvidia. [arxiv]
Exploiting multi-viewpoint consistency for co-training.
LiTS測(cè)試集：95.9， 72.6
- Why could a multiclass dice loss function solve the class imbalance problem?
In cross entropy, each pixel has the same weight irrespective of the class. by using a Dice loss, the weight of a pixel is different. If the CE tumor is small for example, then false positives or false negatives will impact the dice loss more and will thus intrinsically be weighted more.
- New roadmap outlines 5 research priorities for AI in radiology (radiology paper) (healthimaging報(bào)道)
- Novel image reconstruction techniques that quickly produce images humans can read from source data.
- A focus on automated image labeling and annotation, which includes “information extraction from the imaging report, electronic phenotyping and prospective structure image reporting.”
- Machine learning models for clinical data, including pre-trained and distributed learning techniques.
- Algorithms capable of explaining their findings to users.
- Methods for deidentifying images and sharing image datasets that are adequately validated.
FDA developing new rules for artificial intelligence in medicine (news)
Can crowd-sourcing AI algorithms work in radiation oncology? (news) (JAMA paper)
Segmentation models with pretrained backbones (pytorch)
