No copyright over simple colors and shapes used to annotate X-rays

Overjet, Inc. v. VideaHealth, Inc., 2024 WL 3480212, No.24-cv-10446-ADB (D. Mass. Jul. 19, 2024)

The court denied Overjet’s request for a preliminary injunctionon copyright and false advertising claims, concluding that Overjet’s selectionof colors and shapes for annotating dental x-rays was unlikely to be protectedby copyright and that the literally false statements identified by Overjet hadbeen removed, thus preventing a finding of irreparable harm.

“Overjet uses AI to scan dental radiographic X-rays anddetect dental diseases. Specifically, Overjet uses an approach calledsegmentation, which outlines exact borders and shows the extent to whichpathologies, or diseases, exist.” Its software “annotates diseases andstructures on dental X-rays,” which “allows clinicians to better detectdiseases and devise treatment plans.” It’s presently the only dental AI companywith FDA clearance for “performing dental detection and segmentation to outlineexactly where pathologies exist.”

In fall 2023, Overjet launched its “Anatomical StructuresVisualization Tool” for Overjet Caries Assist, which introduced the relevantcoloring scheme and design. The tool uses “white to represent enamel, purple todenote the pulp of a tooth, bright green lines with circles at the ends toindicate measurements of bone levels, and red and translucency to mark areas ofdecay.” [Id. Overjet allegedly chose these colors and shapes in order to createa “contrast with the gray scale” of the X-rays and with the ultimate goal ofassisting with “patient assessment, diagnosis, and treatment planning.” It alsoallegedly “selected the design and colors to help with brand identification;the purple, for example, complements Overjet’s purple brand color.” Exampleimage showing use of red, white, purple, and green lines ending in dots: 

Videa is a competing dental AI company. In December 2023,the FDA cleared “Videa Dental AI,” for “identifying and localizing a range ofdental findings and indications, including caries or cavities and periapicalradiolucency, when analyzing bitewing, periapical, and panoramic X-rays takenfrom patients three years of age and older.” If the software detects anyindications, it “returns a set of bounding boxes representing the suspectdental finding.” Unlike Overjet, VideaAI offers a Clinical View and PatientView; the former “uses boxes and lines with dots to quickly display … points ofinterest.” Patient View, at issue here, is designed to be a “communication toolfor [ ] provider[s].” and offers “a version of the results that are ‘segmented’and highlights the shape of the points of interest” by “show[ing] more thanmere lines and boxes.”

 Example; patient view with red areas, blue, white, and greenish lines with circles:

Videa beat Overjet out of a lucrative partnership with alarge dental client.

Copyright: Overjet’s copyright protected its source code,but there were no allegations of code copying. Instead, it alleged copying of its“coloring scheme, shades, and shape.” First, there were real questions aboutwhether actual copying occurred, given the timeline, which I won’t get into,though those questions are related to the existence of protection for, e.g.,selecting green as the color for healthy gum levels, yellow for some risk,orange for higher risk, and red for a “potential problem.” Videa argued that itchose this “stoplight” system because it was “intuitive for patients and clinicians.”Likewise, it argued that red boxes to indicate tooth decay was an “obviouscolor choice to indicate warning ... and commonly used across industries.” Videafurther argued that it used white to indicate enamel and blue to demarcate pulpwas made “because white is the color of enamel and blue provides contrast torender pulp more visible.”

Second, the color scheme, shades, and shapes were notsufficiently creative/concrete to get copyright protection. The court beganwith 17 U.S.C. § 102(b): “In no case does copyright protection for an originalwork of authorship extend to any idea, procedure, process, system, method ofoperation, concept, principle, or discovery, regardless of the form in which itis described, explained, illustrated, or embodied in such work.”

The court found that Overjet’s claim to shapes, colors, andshading defined the relevant “abstraction” for the infringement test. The nextstep was filtration: eliminating unprotectable elements. Functional elementsare uncopyrightable. The scheme here seemed functional:

The underlying impetus of the colorscheme and display is seemingly to aid clinicians in carrying out dentaldiagnostics.. Thus, although there may not be industry-standards or guidelinesspecifying what colors and shapes may be used on an X-ray, it neverthelessappears intuitive to use certain specific colors to indicate problem areas orcreate contrast in order to efficiently identify pathologies, suggesting thatOverjet’s color, shape, and shading decisions were primarily driven byfunctionality.

Overjet’s own complaint indicated that the tool was “atremendous aid in patient education …. Highlighting enamel and pulp allows fora richer understanding of the nature and severity of disease progression and avisual that matches dentists’ quantified AI finding.”

Relatedly, the individual colors and shades did not rise tothe necessary low level of creativity. Although the “selection, coordination,and arrangement” of elements that are by themselves not protectable maynonetheless meet the originality requirement, that wasn’t the case here. It didnot seem to the court that “countless artistic decisions” were possible, as theywould be for turning a real animal into a stuffed toy. Overjet “organized andarranged only four colors (red, white, purple, and green) and two shapes (aline with circles at the end).”

False advertising:

Overjet challenged statements that allegedly falselyindicated that (1) “Videa’s software is safe and effective for various medicalpurposes,” even though it had not obtained the relevant FDA clearance, and (2) ithad “reached a technological milestone in AI development before Overjet.”

For (1): In a now-deleted LinkedIn post from December 2023, Videastated, “[h]ear from the clinicians who are experiencing firsthand VideaAI’stransformative power.” [The X-ray accompanying the statement allegedly impliedthat Videa had FDA clearance for segmentation when, in reality, the FDA hadonly cleared Videa’s “bounding box” approach. But “[b]ecause Videa is in factan AI platform, the statement is not explicitly false. The Court cannot,however, without the benefit of the full LinkedIn post, assess the extent towhich Videa’s statement, in light of the accompanying X-ray, is misleading, andrefrains from evaluating Statement # 1 under the first prong of the Lanham Actat this juncture.”

In 2021, Videa represented that its AI software can “aid inthe diagnosis and prediction of the progression of diseases such as caries orperiodontitis.” But Videa has not been cleared for progression analysis. Videoargued that it “merely provides this comparative tool to aid clinicians’analyses.” The court found that the challenged statement emphasized thebenefits of “AI within diagnostics and treatment planning,” [Ex. 27 at 2], and wasneither false nor misleading. Its software localizes and identifies variousdental indications in X-rays using bounding boxes. “In comparing theseAI-generated annotated X-rays over time, clinicians can assess the extent towhich certain conditions, including caries and periodontitis, have progressed.”Thus, the statement didn’t communicate that Videa provided progressionanalyses or that Videa had obtained FDA clearance for conducting such analyses.Similar analysis applied to Videa’s statement that its “artificial intelligencesoftware ... helps gauge the caries level in carious lesions.” (The court alsonoted that, when the blog post was published, it contained an “investigational”disclaimer; that disclaimer was removed when Videa obtained the relevant FDAclearance.)

In 2024, Videa blogged that “[w]e suggest choosing an AIprovider with the most FDA-cleared annotations.” According to Videa,annotations are “not equivalent to segmentation,” for which it’s not cleared, butrather refer to “a visual way of showing the results of AI analysis, which cantake many forms, including a simple bounding box.” This wasn’t literally false,because Videa had just obtained FDA clearance for thirty plus indications. Norwas there a necessary implication that “annotation,” that is, labeling anX-ray, refers to FDA clearance for segmentation. That was too attenuated.

Overjet argued that the statement was misleading because theimage right above the alleged misrepresentation is captioned “VideaAI’s patientview showing a segmented x-ray,” which could create an implied connectionbetween “annotations” and “segmentations,” but there was no evidence of actualdeception.

However, another ad, a segmentation view with the caption“[i]ntroducing the first FDA-cleared dental AI for patients aged 3 and up,” whoseimages were subsequently replaced, could lead a viewer to recognize the claim thatVidea obtained FDA clearance for segmentation as “readily as if it had beenexplicitly stated” because of the combination of the image sand text.

Videa also stated that it is the “only dental AI company tospend time and money to retest and improve [caries] models.” Overjet arguedthat it had also done so, submitting a new iteration of its model after the FDAcleared it, and the court agreed that this rendered the statement literallyfalse. Likewise, an initial statement that “[p]roviders can use VideaAI to helpdiagnose all patients, regardless of age” was literally false; it has beenmodified to “[p]roviders can use VideaAI to help diagnose all patients 3 yearsand older.” And another removed statement, “[t]he world’s first pediatricdental AI” was literally false because Overjet obtained clearance for pediatricpatients—at least 12,efore Videa did, even though it was the first to receiveclearance for patients 3 and older. “Pediatric care refers to the treatment ofinfants, children, adolescents, and young adults, and commonly encompasses allages up to twenty-one years old.”

Another statement that VideaAI could “identify cysts” waspotentially false, even though it was FDA-cleared for periapical radiolucency(“PRL”) detection, which includes cysts; identifying a PRL might not be thesame thing as identifying a cyst. So too with a statement that Videa’s software“annotates 18+ months of historical images to help monitor and trackprogression of disease.” The software itself doesn’t track or quantify progression,but seeks to assist clinicians in assessing side by side comparisons ofannotated X-rays. Although the statement was literally true, an image captioned“Current Visit” indicates that the subject tooth’s caries size has increased by20% since the previous X-ray was taken. That image, coupled with the statementthat Videa’s “AI annotates 18+ months of historical images to ... trackprogression of disease,” could suggest that the software—rather than theclinician—evaluates the progression.

Of the surviving statements, there was “scant” support formateriality. Statements focusing on the idea that Videa’s AI software can aidin the diagnosis and prediction of various diseases didn’t seem material. Butreferences to VideaAI’s pediatric qualities were more plausibly material. Nonetheless,Videa sells to sophisticated customers such as dental practices and services,who “likely have the scientific and technological understanding to fullyappreciate Videa’s product offering and its limitations.”

Online advertising andpresentations during trade shows appear to be a starting point for potentialpurchases, but do not alone drive such decisions. It is reasonable to assumethat, given the actors involved and the considerable monetary value of potentialcontracts, the sales process is thorough and lengthy, with a heavy focus on thesoftware’s actual capabilities, including relevant FDA clearance.

What about bait and switch reasoning?

There was also no evidence of actual deception. “It seemsreasonable that discussions surrounding relevant FDA clearance—and verificationthereof—are a critical component of the sales process. A few promotional videosand blog posts, it appears, are unlikely to cause consumer deception among therelevant pool of potential consumers.”

Nor did Overjet show injury from the deception. This alsoprevented a showing of irreparable harm, especially as the literally falsestatements have been removed.