Confocal video-mosaicking microscopy to guide surgery of superficially spreading skin cancers

表面播散型皮肤癌，如恶性黑色素瘤(LMM)和非黑色素瘤 皮肤癌(NMSCs)主要发生在老年患者，弥漫性亚临床皮肤癌广泛分布于 区域和边界界定不清，很难发现。为了治疗这些癌症，皮肤科医生-- 定期进行大量标测活组织检查以确定扩散和边缘，然后进行手术 切除时有很大的“安全”边际。毫不奇怪，这种“盲目”的方法会导致对 边缘，正常皮肤的过度采样，太多的假阳性和假阴性，以及太多的损失 正常皮肤组织。可以帮助解决这个问题的是反射共焦显微镜(RCM)成像 非侵入性地描绘边缘，直接在患者身上。RCM成像对体内皮肤癌的检测具有敏感性 85-95%，特异性80-70%。2016年，医疗保险和医疗补助服务中心批准了Reim- 皮肤RCM成像的颜色代码。RCM成像现在越来越多地被用于无创 指导诊断，避免患者进行不必要的良性病变活检。虽然二十年来的努力 导致授予这些代码的重点是影像引导诊断，新兴的应用是在 影像指导治疗。我们建议创建一种称为RCM视频马赛克的方法，以非侵入性 绘制患者大范围皮肤癌边缘图，增加采样、准确性和节省NOR- MAL组织。创新之处在于设计了一种高度健壮的(防止组织扭曲和运动伪影) 和高速(实时，秒)的RCM视频马赛克方法：我们将开发一种光流AP- 由8个参数组成的新型混合三阶段深度学习网络将对全局进行建模 和局部刚性和非刚性组织运动动力学，学习和适应变化的组织和斑点噪声。 并预测和自动检测运动模糊伪影。根据PAR-18-009的要求，我们的 学术-产业合作伙伴关系将向临床医生提供RCM视频马赛克，以便在 床边(翻译新奇)。我们建议的应用是为了指导手术切除，但该方法 将产生更广泛的影响，指导新的和正在出现的侵入性较小的浅表皮肤非手术治疗 癌症。在一项初步研究中，我们演示了实时速度(125毫秒- 秒/帧、8帧/秒)，并且相对于视场的配准误差为1.02±1.3像素 1000 x 1000像素。我们的具体目标是(1)开发一个实时和健壮的RCM视频马赛克AP- 在床边使用的手持式共聚焦显微镜，(2)测试这种方法 图像质量和临床可接受性，以及(3)对100名患者进行前瞻性测试，术前视频- 对LMM切缘和浅层NMSC切缘进行镶嵌，然后对照术后检查进行验证。 这就是问题。我们是来自东北大学纪念斯隆·凯特琳癌症中心的一个高度协同的团队- Verthy，以及口径成像和诊断公司(以前是Lucid Inc.)，有13年的合作记录。

Superficially spreading types of skin cancers such as lentigo maligna melanomas (LMMs) and non-melanoma skin cancers (NMSCs) occur mostly on older patients, with diffuse sub-clinical sub-surface spread over large areas and with poorly defined margins that are difficult to detect. To treat these cancers, dermatologists rou- tinely perform a large number of mapping biopsies to determine the spread and margins, followed by surgical excision with wide "safety" margins. Not surprisingly, such a "blind" approach results in under-sampling of the margins, over-sampling of normal skin, too many false positives and false negatives, and too much loss of normal skin tissue. What may help address this problem is reflectance confocal microscopy (RCM) imaging to noninvasively delineate margins, directly on patients. RCM imaging detects skin cancers in vivo with sensitivity of 85-95% and specificity 80-70%. In 2016, the Centers for Medicare and Medicaid Services granted reim- bursement codes for RCM imaging of skin. RCM imaging is now being increasingly used to noninvasively guide diagnosis, sparing patients from unnecessary biopsies of benign lesions. While the two-decade effort leading to the granting of these codes was focused on imaging-guided diagnosis, emerging applications are in imaging to guide therapy. We propose to create an approach called RCM video-mosaicking, to noninvasively map skin cancer margins over large areas on patients, with increased sampling, accuracy and sparing of nor- mal tissue. The innovation will be in designing a highly robust (against tissue warping and motion artifacts) and high speed (real-time, seconds) approach for RCM video-mosaicking: we will develop an optical flow ap- proach with a novel hybrid 3-stage deep learning network comprising of 8 parameters that will model global and local rigid and non-rigid tissue motion dynamics, learn and adapt to variable tissue and speckle noise con- ditions in patients, and predict and automatically detect motion blur artifacts. As required by PAR-18-009, our academic-industrial partnership will deliver RCM video-mosaicking to clinicians for real-time implementation at the bedside (translational novelty). Our proposed application is for guiding surgical excision, but the approach will have wider impact, for guiding new and emerging less invasive non-surgical treatments for superficial skin cancers. In a preliminary study, we demonstrated RCM video-mosaicking with real-time speed (125 millisec- onds per frame, 8 frames per second), and registration errors of 1.02 ± 1.3 pixels relative to field-of-view of 1000 x 1000 pixels. Our specific aims are (1) to develop a real-time and robust RCM video-mosaicking ap- proach and incorporate into a handheld confocal microscope for use at the bedside, (2) to test the approach for image quality and clinical acceptability, and (3) to prospectively test on 100 patients, with pre-surgical video- mosaicking of LMM margins and superficial NMSC margins, followed by validation against post-surgical pa- thology. We are a highly synergistic team from Memorial Sloan Kettering Cancer Center, Northeastern Uni- versity, and Caliber Imaging and Diagnostics (formerly, Lucid Inc.), with a 13-year record of collaboration.