Accurate diagnosis of thyroid follicular lesions from nuclear morphology using supervised learning

Abstract

Follicular lesions of the thyroid remain significant diagnostic challenges in surgical pathology and cytology. The diagnosis often requires considerable resources and ancillary tests including immunohistochemistry, molecular studies, and expert consultation. Visual analyses of nuclear morphological features, generally speaking, have not been helpful in distinguishing this group of lesions. Here we describe a method for distinguishing between follicular lesions of the thyroid based on nuclear morphology. The method utilizes an optimal transport-based linear embedding for segmented nuclei, together with an adaptation of existing classification methods. We show the method outputs assignments (classification results) which are near perfectly correlated with the clinical diagnosis of several lesion types' lesions utilizing a database of 94 patients in total. Experimental comparisons also show the new method can significantly outperform standard numerical feature-type methods in terms of agreement with the clinical diagnosis gold standard. In addition, the new method could potentially be used to derive insights into biologically meaningful nuclear morphology differences in these lesions. Our methods could be incorporated into a tool for pathologists to aid in distinguishing between follicular lesions of the thyroid. In addition, these results could potentially provide nuclear morphological correlates of biological behavior and reduce health care costs by decreasing histotechnician and pathologist time and obviating the need for ancillary testing.

Automated classification of brain tumor type in whole-slide digital pathology images using local representative tiles

“…Accuracy of the model is stable over much of the parameter range, with lower accuracy only occurring where lambda forces few to no features to be included in the model. Dundar et al, 2011Dundar et al, , 2010Esgiar et al, 2002;Farjam et al, 2007;Fatima et al, 2014;Glotsos et al, 2008;Gunduz et al, 2004;Gupta et al, 2001;Hall et al, 2008;Huang and Lai, 2010;Huang and Lee, 2009;Jafari-Khouzani and Soltanian-Zadeh, 2003;Monaco and Tomaszewski, 2008;Kong et al, 2007;Kwak et al, 2011;Land et al, 2008;Lessmann et al, 2007;Meng et al, 2010;Naik et al, 2008Naik et al, , 2007Ozolek et al, 2014;Petushi et al, 2006;Qureshi et al, 2008Qureshi et al, , 2007Sboner et al, 2003;Schnorrenberg et al, 1997;Sertel et al, 2010;Sparks and Madabhushi, 2013;Sudbø et al, 2000;Tabesh and Teverovskiy, 2006;Tabesh et al, 2007Tabesh et al, , 2005Tahir and Bouridane, 2006;Tasoulis et al, 2003;Teverovskiy et al, 2004;Thiran and Macq, 1996;Tsai et al, 2009;Xu et al 2014aXu et al , 2014bYang et al, 2009). These approache...…”

“…Many other methods collect features from either the original RGB image, a converted image to other non-biologically based color spaces (e.g., Lab or HSL), or from a grayscale version of that same image (Al-Kadi, 2010;Basavanhally et al, 2010;Dundar et al, 2011Dundar et al, , 2010Esgiar et al, 2002;Farjam et al, 2007;Glotsos et al, 2008;Huang and Lee, 2009;Jafari-Khouzani and SoltanianZadeh, 2003;Kong et al, 2009;Ozolek et al, 2014;Petushi et al, 2006;Qureshi et al, 2008;Ruiz et al, 2007;Schnorrenberg et al, 1997;Tabesh et al, 2005;Tabesh and Teverovskiy, 2006;Tahir and Bouridane, 2006;Thiran and Macq, 1996;Tuzel et al, 2007;Wang et al, 2010;Wetzel et al, 1999;Weyn et al, 1998;Xu et al 2014aXu et al , 2014b. Since hematoxylin binds to nucleic acids and eosin binds to protein, unmixing the stains allows the feature extraction to directly probe the state of these important biological molecules, whereas features from the mixed image may either miss this signal or be unable to probe them independently.…”

“…Even before WSI was practical, researchers captured a single frame of an entire slide using a standard microscope creating a subset of the image for computational analysis (Adiga et al, 2006;Aiad et al, 2009;Altunbay et al, 2010;Baak et al, 1981;Brook et al, 2007;Esgiar et al, 2002Esgiar et al, , 2002Farjam et al, 2007;Fatima et al, 2014;Glotsos et al, 2008;Gunduz et al, 2004;Gupta et al, 2001;Hall et al, 2008;Huang and Lai, 2010;Huang and Lee, 2009;Jafari-Khouzani and Soltanian-Zadeh, 2003;Kong et al, 2007;Land et al, 2008;Lessmann et al, 2007;Meng et al, 2010;Ozolek et al, 2014;Qureshi et al, 2008Qureshi et al, , 2007Sboner et al, 2003;Schnorrenberg et al, 1997;Sudbø et al, 2000;Tahir and Bouridane, 2006;Tasoulis et al, 2003;Thiran and Macq, 1996;Tsai et al, 2009). For many researchers doing pathology image analysis today, this is still the best option, either because they do not have access to a WSI system, or because they can capture data not possible with a WSI system, such as confocal microscopy images.…”

A continuous linear optimal transport approach for pattern analysis in image datasets

“…One such well-suited application is the detection of cancer from cell morphology. We have recently shown that the discrete LOT technique can be used to classify certain types of thyroid cancers with near perfect sensitivity and specificity [25] on a cohort of 94 patients. At the same time the technique can be utilized to provide biological information regarding chromatin reorganization differences in different cancer types.…”

Label-free identification of non-activated lymphocytes using three-dimensional refractive index tomography and machine learning

“…This approach is especially powerful for high-dimensional data that are extremely difficult to manually process by a human due to the complicity and large size 38 . In biomedical research fields, machine learning methods have been widely used to solve complex biological problems: identification of bacterial species 39 , discrimination of WBC types 40,41 , and investigation of pathophysiologic conditions [42][43][44] . However, there has been no applications that use both 3-D RI tomography and machine learning for the purpose of biomedical studies.…”